南京信息工程大学学报
南京信息工程大學學報
남경신식공정대학학보
JOURNAL OF NANJING UNIVERSITY OF INFORMATION SCIENCE & TECHNOLOGY
2014年
5期
449-458
,共10页
统计降尺度%日降水%BP 神经网络%旋转主成分分析(REOF)
統計降呎度%日降水%BP 神經網絡%鏇轉主成分分析(REOF)
통계강척도%일강수%BP 신경망락%선전주성분분석(REOF)
statistical downscaling%daily precipitation%BP neural network model%Rotated Empirical Orthogonal Function(REOF)
利用1999-2009年安徽省淮河以南地区60个县市站夏季逐日降水资料和安庆市探空站逐日资料，研究了中低层不同风向配置下局地降水与大尺度降水场之间的关系，以3种不同预报对象及相应的预报因子分别采用神经网络和线性回归方法设计6种预报模型对观测资料进行逼近和优化，从而实现空间降尺度。分析对比6种预报模型46站逐日降水量的拟合和预报效果，结果表明：采取相同的预报对象及预报因子的BP 神经网络模型在拟合和预报效果上均好于线性回归模型，可见夏季降水场之间以非线性相关为主；神经网络模型预报结果同常用的Cressman插值预报相比，能很好地反映出降水的基本分布及局地特征；预报对象为单站降水序列的神经网络模型在以平原、河流为主要地形的区域预报效果较好，预报对象为 REOF主成分的神经网络模型则在山地和丘陵地形区域预报效果较好。
利用1999-2009年安徽省淮河以南地區60箇縣市站夏季逐日降水資料和安慶市探空站逐日資料，研究瞭中低層不同風嚮配置下跼地降水與大呎度降水場之間的關繫，以3種不同預報對象及相應的預報因子分彆採用神經網絡和線性迴歸方法設計6種預報模型對觀測資料進行逼近和優化，從而實現空間降呎度。分析對比6種預報模型46站逐日降水量的擬閤和預報效果，結果錶明：採取相同的預報對象及預報因子的BP 神經網絡模型在擬閤和預報效果上均好于線性迴歸模型，可見夏季降水場之間以非線性相關為主；神經網絡模型預報結果同常用的Cressman插值預報相比，能很好地反映齣降水的基本分佈及跼地特徵；預報對象為單站降水序列的神經網絡模型在以平原、河流為主要地形的區域預報效果較好，預報對象為 REOF主成分的神經網絡模型則在山地和丘陵地形區域預報效果較好。
이용1999-2009년안휘성회하이남지구60개현시참하계축일강수자료화안경시탐공참축일자료，연구료중저층불동풍향배치하국지강수여대척도강수장지간적관계，이3충불동예보대상급상응적예보인자분별채용신경망락화선성회귀방법설계6충예보모형대관측자료진행핍근화우화，종이실현공간강척도。분석대비6충예보모형46참축일강수량적의합화예보효과，결과표명：채취상동적예보대상급예보인자적BP 신경망락모형재의합화예보효과상균호우선성회귀모형，가견하계강수장지간이비선성상관위주；신경망락모형예보결과동상용적Cressman삽치예보상비，능흔호지반영출강수적기본분포급국지특정；예보대상위단참강수서렬적신경망락모형재이평원、하류위주요지형적구역예보효과교호，예보대상위 REOF주성분적신경망락모형칙재산지화구릉지형구역예보효과교호。
Based on the summer daily precipitation data of 60 meteorological stations in Anhui province from 1999 to 2009 and observation data of Anqing sounding station,the relationship between local precipitation and large-scale precipitation circulation in different mid-low wind directions are studied in this paper.The neural network and linear regression method,combined with 3 forecasting objects and corresponding predictor variables are employed to design 6 downscaling function models to approximate and optimize the precipitation data.The 6 models are used to simulate and forecast the daily precipitation data of 46 meteorological stations in Anhui province,and the results show that BP neural network models generally outperform the linear regression models in simulation and forecasting accuracy, indicating the nonlinear correlation between different scales of summer rainfall. Compared with the commonly used Cressman interpolation methods,the neural network models can reflect the basic distribution and local characteristics of summer precipitation in forecasting results.The BP neural network model with single station precipitation series as prediction object has good forecasting results in areas of plains or rivers,while the BP neural network model with the REOF principal components as predicting object is good in mountainous area.
