BP人工神经网络在MM5预报福建沿海大风中的释用  

Interpretation and Application of BP Artificial Neural Network in MM5 Model Forecasting Gale for Coastal Regions of Fujian Province

作　　者：陈德花 刘铭[2] 苏卫东 夏丽花[2] 石顺吉 

机构地区：[1]厦门市气象台,厦门361012 [2]福建省气象台,福州353001

出　　处：《暴雨灾害》2010年第3期263-267,共5页Torrential Rain and Disasters

基　　金：福建省自然科学基金计划项目(W0650004);厦门市科技局科技计划项目(3502Z20064022);"厦金航线气象保障项目"

摘　　要：利用MM5中尺度数值模式输出的福建沿海6个气象站2004年5月到2007年10月每天08、20时48 h每6 h间隔的风速预报和实况资料,采用不同隐层以及节点数,按照风速大小分类建立人工神经网络模型,以此为基础应用BP人工神经网络建立风速预报模型,并将该模型应用到2008年1月至2009年2月福建沿海平潭、崇武、东山三站风力预报,对其效果进行检验。结果表明,采用一层隐层3个隐层节点数的人工神经网络模型是预报风速的最佳模型;经人工神经网络订正后,沿海风速预报比MM5模式预报有很大改善,特别是对大风(>10 m.s-1)预报能力有极大提高,其Vs评分比MM5模式提高60分;经检验,经人工神经网络订正后的风速预报精度比MM5模式提高约32.3分,总体上,随风力增大,订正后的风速预报效果越好。Based on BP artificial neural network method,the wind speed observation data from the 6 coastal weather stations in Fujian in the period from May 2004 to October 2007 and 6-hourly interval wind speed forecast from MM5 meso-scale numerical model in view of the same 6 coastal weather stations and the same period were used to analyze the interpretation and application of gale forecast by MM5.The wind speed was used to build the artificial neural network models with different hidden layer and nodes.These models were used to test wind speed forecast by Pingtan,Chongwu and Dongshan stations in the coastal regions of Fujian province in the period form January 2008 to February 2009.The results show that the model using 1 hidden layer with 3 hidden nodes is the best one to forecast wind speed.After corrected by artificial neural network model,the coast wind forecast accuracy is greatly improved comparing with the forecast by MM5.The forecast capability of gale（10 m.s-1） is greatly increased,and Vs score of the models is raised by 60 points over MM5.By inspection,the wind speed forecasting accuracy is increased by about 32.3 points after corrected by artificial neural network model.Generally,with wind speed faster,the corrected wind speed forecasting results are better.

关 键 词：MM5中尺度数值模式 BP人工神经网络 风速订正 风速预报 

分 类 号：P458.123[天文地球—大气科学及气象学]