基于k-最近邻筛选的BMA集合预报模型研究

发布时间：2018-10-20 14:59

【摘要】：针对冗余训练样本会降低BMA参数求解效率与精度问题,本文提出在BMA运算之前采用k-最近邻(k-nearest neighbor)算法筛选有价值训练样本,并用于BMA参数求解的改进模型。模拟试验在淮河王家坝站进行,分别以k-最近邻筛选、不筛选两种方案为BMA提供训练样本,统计分析两种方案中王家坝站流量模拟结果,评价BMA改进法的性能。模拟结果显示,采用k-最近邻样本筛选方法后,BMA模型对洪水过程以及洪峰的预报精度提升明显;概率预报结果的离散程度降低的同时,可靠性程度获得提升。k-最近邻样本筛选方法的引入,能够有效去除BMA模型训练样本中的冗余数据,以少量的样本获得更可靠的模型参数,改善集合预报性能。
[Abstract]:In view of the problem that redundant training samples can reduce the efficiency and precision of BMA parameter solving, this paper presents an improved model of selecting valuable training samples by using k- nearest neighbor (k-nearest neighbor) algorithm before BMA operation, which can be used to solve BMA parameters. The simulation test was carried out at the Wangjiaba Station of Huaihe River. The two schemes were screened by the nearest neighbor, and the two schemes were not screened to provide training samples for the BMA. The flow simulation results of the Wangjiaba Station in the two schemes were statistically analyzed, and the performance of the improved BMA method was evaluated. The simulation results show that the accuracy of BMA model for flood process and Hong Feng is improved obviously, and the dispersion of probabilistic forecast results is reduced. The introduction of K-nearest neighbor sample screening method can effectively remove redundant data from training samples of BMA model, obtain more reliable model parameters with a small number of samples, and improve the performance of ensemble prediction.
【作者单位】：淮河水利委员会水文局(信息中心);河海大学水文水资源学院;
【基金】：国家重点研发计划项目(2016YFC0400909) 国家自然科学基金项目(41130639,51179045,41101017,41201028)
【分类号】：P338

【相似文献】