基于改进径向基函数的降雨量短期预测研究  

Short-term Rainfall Forecasting Based on a Modified RBF Function

作　　者：董淑华 邢贞相[2] 娄丹 张玉国 张涵[2] 郭皓[2] DONG Shu-hua;XING Zhen-xiang;LOU Dan;ZHANG Yu-guo;ZHANG Han;GUO Hao(Hydrological Bureau of Heilongjiang Province,Harbin 150001,China;School of Water Conservancy & Civil Engineering,Northeast Agricultural University,Harbin 150030,China)

机构地区：[1]黑龙江省水文局,哈尔滨150001 [2]东北农业大学水利与土木工程学院,哈尔滨150030

出　　处：《沈阳农业大学学报》2017年第3期367-372,共6页Journal of Shenyang Agricultural University

基　　金：国家自然科学基金项目(51109036;51179032);黑龙江省自然科学基金项目(E2015024);教育部高等学校博士点学科专项科研基金项目(20112325120009);黑龙江省级领军人才梯队后备带头人资助项目(500001);黑龙江省水利厅科技开发项目(201402;201403);东北农业大学"学术骨干"项目(518536)

摘　　要：为提高短期降水量预测的精度,尤其是汛期降水量的准确估计对防洪减灾以及水资源管理都具有很重要的指导意义。将具有较强非线性映射能力的人工神经网络技术用于汛期降水量预测,更符合降水量的随机相关特征,能切实提高其预测精度。将基于密度参数改进传统K-均值算法与径向基函数神经网络(radial basis function,RBF)耦合,提出了一种新的短期降雨量预报模型。并将所构建的模型应用于黑龙江省双鸭山市汛期月降雨量预报中进行验证。RBF神经网络是一种泛化能力较强的前馈型神经网络模型;密度参数可以通过寻找聚类中心至样本平均距离区域内样本的最优密度,使K-均值算法确定的RBF神经网络基函数中心减少波动,消除标准K-均值算法对初始聚类中心的敏感性,提高RBF网络的逼近能力和网络中心的搜索速度。研究结果表明:基于改进径向基函数的降雨预报模型对于预见期2010年、2011年和2012年的汛期(6~9月)降雨量的计算平均相对误差为10.81%,确定性系数为0.95,预报精度比标准K-均值算法和BP神经网络两种模型的计算结果都有所提高。本研究确定的径向基函数能够更好地描述研究区域汛期月降雨量间的映射关系,与标准K-均值算法和BP神经网络两种模型相比,除预报精度有所提高外,其收敛速度更快,这表明本方法能对短期降水预报提供更高的预报精度。In order to improve the accuracy o f sh o rt-term precipitation forecasting,especially for the forecasting of rainfall inflood season,a monthly rainfall forecasting model of radial basis function(RBF)based on a K-m ean s algorithm modified by adensity param eter was built in this study.The RBF neural network is an artificial neural network with strong generalizationability,which is more suitable to stochastic correlation characteristics of rainfall.K-m ean s algorithm modified with the densityparam eter,which can find the optimal density of the sam ples in the area between the cluster center and the m ean distance ofsamples,and reduce fluctuation and sensitivity of initial clustering centers in a standard K-m eans algorithm.The K-m eansalgorithm based on the density param eter could improve the approximation capability and the searching speed of the RBF ANNcenter.The proposed model was used to forecast monthly rainfall in Shuangyashan area during the flood season.The forecastresults of the proposed model were com pared with that of the standard K-m ean RBF ANN model and Back propagation ANNmodel.The case study showed that the relative m ean error of rainfall forecasting in flood season(from June to September)of theyear2010,2011and2012was10.81%,and the determ inistic coefficient was0.95.The calculation accuracy of the proposedmodel was better than that of the standard K-m eans RBF ANN model and BP ANN model.The RBF based K-m eans modifiedby density param eter could reflect the relationship between monthly rainfalls in flood season.Moreover,the method had fastconvergence speed and rem arkable accuracy.

关 键 词：径向基函数 密度参数 K-均值 汛期降雨量 

分 类 号：P338[天文地球—水文科学]