摘要
利用1993-2011年逐日地面资料对福建省18个地面站的雾日数进行了统计分析,且利用了2007-2012年1、2、11和12月的高空及T213数值预报资料选取预报因子,进行了相关分析,并分别建立了逐步回归能见度模型、BP神经网络能见度模型和判别大雾3种模型。结论如下:(1)从回代检验结果中看出BP网络模型稳定性次于逐步回归模型,但是BP网络模型的预测精度和成功率均高于逐步回归模型;(2)在试预报时候BP网络模型的预报效果较逐步回归模型更优秀,3个模型中判别大雾的模型准确率最高。建议在日常预报中使用BP网络大雾判别预报模型。
This article statistically analyzed the fog days of 18 earth stations in FUJIAN province by the day to day ground data from Jan to Dec of year 1993 to 2011 and selected predictors, basing on the upper air data in Jan, Feb, Nov & Dec of year 2007 to 2012 and T213 material to establish three fog forecasting models, they are Visibility Stepwise regression, BP feedforward neural network visibility and Judging Thick Fog. Key findings are as following,(1)BP Feedforward neural network works better in prediction accuracy and success rate during back substitution test while stability falls behind stepwise regression method.(2)In trial forecast, BP feedforward neural network has a better forecasting result than stepwise regression.Judging thick fog is the most accurate model among them with a minimum mean square error, so we suggest to applying BP feedforward neural network thick fog judging model in daily fog forecasting.
出处
《环境科学与技术》
CAS
CSCD
北大核心
2015年第S2期404-407,共4页
Environmental Science & Technology
关键词
大雾
预报因子
双线性插值
逐步回归
BP神经网络
fog
predictor
Bilinear interpolation
stepwise regression
BP Feedforward neural network