KNN数据挖掘算法在北京地区霾等级预报中的应用  

Application of KNN Data Mining Algorithm to Haze Grade Forecasting in Beijing

作　　者：熊亚军[1,2] 廖晓农[1,2] 李梓铭[1,2] 张小玲[1,2] 孙兆彬[1,2] 赵秀娟[1,2] 赵普生[1,2] 马小会[1,2] 蒲维维[1,2] 

机构地区：[1]京津冀环境气象预报预警中心,北京100089 [2]中国气象局北京城市气象研究所,北京100089

出　　处：《气象》2015年第1期98-104,共7页Meteorological Monthly

基　　金：中央级公益性科研院所专项基金(IUMKY201303PP0103);国家科技支撑计划项目(2014ABC16B04);北京市科技计划项目(Z131100006113013);首都蓝天行动培育专项(Z141100001014013)共同资助

摘　　要：利用北京地区2013年气象数据以及PM_(2.5)浓度数据与能见度数据进行对比分析,结果发现气温、气压、相对湿度、露点温度、地面U风、地面V风以及PM_(2.5)小时浓度这7个要素是影响北京地区霾等级的关键因素。利用气温、地面气压、相对湿度、露点温度、U风、V风分量以及PM_(2.5)浓度作为7个属性特征,以霾等级做为标志量构建训练样本集,结合KNN(KNearest Neighbor)数据挖掘算法构建疆等级预报分类器,并开展霾等级客观识别实验。结果表明K=3时该分类器的分类预报效果最佳,其13个站点的分类准确率高达88.2%。基于该算法构建的KNN模型预报无霾时的漏报概率很小,准确率高达91.8%;预报有轻度霾、中度霾以及重度霾时,空报的概率仅分别为4.7%、1.4%和2.6%。2014年8月29日至9月2日北京地区一次霾天气过程的预报结果表明:南郊观象台、密云和延庆3站的预报准确率分别达到74%、64%和84%,但霾等级的精度方面还有待于进一步提高。;Based on meteorological data and PM_(2.5) concentration data of 2013,the statistical analysis shows that temperature,pressure,relative humidity,dew point temperature,surface wind,U wind,V wind and PM_(2.5) hourly concentration are the 7 key factors affecting the haze grade of Beijing Area.The attribute vectors of KNN training sample set are the above 7 factors and the label vector is haze grade.The classifier can be established combining KNN Data Mining Algorithm and its best value is parameter K=3.The classification accuracy of the 13 meteorological stations is 88.2%.The forecasting model based on KNN Algorithm has good accuracy.When it predicts no haze,the accuracy rate is 91.8%and missing forecast chance is very small.When mild,moderate or severe haze is predicted,the probability of empty forecast is only 4.7%,1.4%and 2.6%,respectively.There was one froghaze weather process from 29 August to 2 September 2014 in Beijing Area.Its prediction results show the forecast accuracy of Nanjiao,Miyun and Yanqing is 74%,64%and 84%respectively,but the accuracy of the haze grade remains to be further improved.

关 键 词：数据挖掘 KNN 霾 预报 

分 类 号：P456[天文地球—大气科学及气象学] X513[环境科学与工程—环境工程]