摘要
将BP神经网络与K-最近邻(KNN)算法耦合起来,建立BK(BP-KNN)模型,该模型以前期模拟流量和相应影响要素作为BP神经网络的输入,出口断面流量作为网络输出,对产汇流过程进行模拟;采用K-最近邻算法,基于历史样本的模拟误差和相应影响要素对网络输出进行修正,实现了非实时校正模式下的连续模拟。根据BK模型的计算流程将其参数分为3个层次,各层次分别使用NSGA-Ⅱ多目标优化算法进行参数优选,提高了模拟精度、优化效率和网络泛化能力。分别将新安江模型的产流、产流分水源计算模块与BK模型相耦合,建立XBK(Xinanjiang runoff production-BK)和XSBK(Xinanjiang runoff production and separation-BK)模型,在呈村等3个不同类型的流域应用新安江模型、BK模型、XBK模型和XSBK模型进行模拟精度比较,结果表明改进的模型模拟精度更高,较好地解决了神经网络模型在水文模拟中存在的问题。
The BK（BP-KNN） model,which is coupled by the BP neural network model and the K-nearest neighbor（KNN） model,was established.This model was used to simulate flow concentration and runoff generation with antecedent simulated outlet flow and relevant influencing factors as inputs of the BP neural network,and the outlet flow as the output of the network.The output of the network was corrected by the K-nearest neighbor algorithm based on historical samples＇ simulation error and relevant influencing factors.The KNN algorithm realizes continuous simulation in a non-real time mode.According to the computational procedure,parameters of the BK model were divided into three groups and optimized by the NSGA-II multi-object algorithm in each group.This way of calibration improves the accuracy,efficiency,and generalization ability of the BK model.The XBK（Xin＇anjiang runoff production-BK） and the XSBK（Xin＇anjiang runoff production and separation-BK） model were coupled by the Xin＇anjiang runoff production module,the Xin＇anjiang runoff production and separation module,and the BK flow concentration module.BK,XBK,XSBK,and the Xin＇anjiang model were applied to the Chengcun,Dongwan,and Dage watersheds to compare their simulation accuracies.The results show that the improved model has higher accuracy and can be used to solve problems of neural network models during hydrological simulation.
出处
《河海大学学报（自然科学版）》
CAS
CSCD
北大核心
2013年第4期294-299,共6页
Journal of Hohai University(Natural Sciences)
基金
国家自然科学重点基金(41130639)
江苏省普通高校研究生科研创新计划(CXZZ11-0435)
