基于信息熵误差异分布模型的洪水概率预报方法  

Flood probability forecast method based on information entropy error hetero-distribution model

作　　者：毕铭佐 梁忠民[1] 陈在妮 李佳 胡义明[1] BI Mingzuo;LIANG Zhongmin;CHEN Zaini;LI Jia;HU Yiming(College of Hydrology and Water Resources,Hohai University,Nanjing 210098,China;Dadu River Basin Cascade Power Slalion Centralized Control Center,National Energy Group,Chengdu 610041,China)

机构地区：[1]河海大学水文水资源学院,江苏南京210098 [2]国家能源集团大渡河流域梯级电站集控中心,四川成都610041

出　　处：《武汉大学学报（工学版）》2021年第1期58-64,共7页Engineering Journal of Wuhan University

基　　金：国家重点研发计划项目(编号:2016YFC0402709、2018YFC0407206);国电大渡河流域水电开发有限公司科技项目(编号:PDP-KY-2019-001)。

摘　　要：以大渡河流域大金站为对象,开展了基于信息熵误差异分布模型的洪水概率预报方法研究。挖掘了确定性水文模型预报的相对误差的统计规律,发现不同量级流量下洪水预报的相对误差存在异分布特征,服从不同的概率分布函数。基于极大熵和极小熵原则确定不同流量量级下预报相对误差的最优概率分布函数,进而采用不同的误差分布函数描述不同流量量级下的预报误差规律,并推导了以确定性模型预报值为条件的"真实"流量的分量级条件概率分布函数。根据不同量级对应的条件概率分布函数,可获得概率预报倾向值(如中位数预报)及任意置信度下的概率预报区间(如90%预报区间)。大金站的应用结果表明,概率预报模型提供的预报倾向值(中位数预报值)整体要优于原始的确定性模型预报值,同时概率预报模型提供的90%置信度下的预报区间可以涵盖绝大部分实测值,具有较高的可靠性。For the flow forecast of the Dajin Station in Dadu River Basin,the flood probability prediction method based on information entropy error hetero-distribution model is studied.The statistical law of the relative error of the deterministic hydrological model prediction is excavated.It is found that the relative error of the flood forecast of different magnitude of flow has different distribution characteristics,and obeys different probability distribution functions.Based on the principle of maximum entropy and minimum entropy,the optimal probability distribution function of prediction relative error for different magnitude of flow is determined.Then,different error distribution functions are used to describe the prediction error rule for different magnitude of flow.The conditional probability distribution function of"real"flow of different magnitude is derived based on the predicted value of deterministic model.According to the conditional probability distribution function corresponding to different magnitude of flow,the probability prediction preferred value(such as median)and the probability prediction interval(such as 90%prediction interval)under any confidence level can be obtained.The application results of Dajin Station show that the prediction preferred value(median)provided by the probability prediction model is better than that of the original deterministic model,and the prediction interval provided by the probability prediction model under 90%confidence level can cover most of the measured values with high reliability.

关 键 词：洪水概率预报 信息熵 误差异分布 预报倾向值 预报区间 

分 类 号：TV124[水利工程—水文学及水资源]