摘要
以实测样本隶属于某一模型的后验概率为权重,对各模型预报变量的后验分布进行加权平均,获得综合预报变量的概率密度函数,进而推导出均值、方差公式和置信区间。将“基于BP神经网络和遗传算法的分类洪水预报”与“基于贝叶斯模型平均法的多模型组合预报”相结合,实现“洪水分类多模型加权组合预报”。该方法综合考虑了各模型对不同类型洪水的适应条件及多个模型的优势,可同步降低模型参数与模型结构不确定性对预报结果的影响。以大伙房水库为研究区域,采用大伙房水库1960年~2016年的25场历史典型洪水作为本研究的基础资料,分别完成实例验证与分析,效果整体上优于单个模型。
Taking the posterior probability of the measured samples belonging to a certain model as the weight,the posterior distribution of the predicted variables of each model is weighted averaged,and the probability density function of the comprehensive predicted variables is obtained,and then the mean,variance formula and confidence interval are deduced.Combining the"Classified Flood Forecasting Based on BP Neural Network and Genetic Algorithms"with"Multi-model Combination Forecasting Based on Bayesian Model Average Method",this method realizes the"Flood Classification Multi-model Weighted Combination Forecasting",which takes into account the adaptability of each model to different types of floods and the advantages of multiple models,and can be synchronized to reduce the influences of model parameters and model structure uncertainty on the prediction results.Taking Dahuofang Reservoir Basin as the research area,25 typical historical floods of Dahuofang Reservoir from 1960 to 2016 are used as the basic data of this study,and the case validation and analysis are completed respectively.It shows that the flood classification multi-model weighted combination forecasting is better than the results of single model on the whole.
作者
刘恒
LIU Heng(Liaoning Water Conservancy and Hydropower Survey and Design Research Institute Co.,Ltd.,Shenyang 110003,Liaoning,China)
出处
《水力发电》
北大核心
2020年第3期13-20,122,共9页
Water Power
基金
水利部公益性行业科研专项经费项目(200801040)
水利部科技推广计划项目(TG1142)
辽宁省农业攻关计划项目(2011216001)。
关键词
BP神经网络
遗传算法
贝叶斯平均法
分类组合预报
大伙房水库
BP neural network
genetic algorithms
Bayesian average method
classified combination forecasting
Dahuofang Reservoir
