本文已被：浏览 4555次 下载 22498次
投稿时间：2014-09-04 修订日期：2014-11-18
投稿时间：2014-09-04 修订日期：2014-11-18
中文摘要: 利用北京地区2013年气象数据以及PM2.5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PM2.5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PM2.5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（K Nearest Neighbor）数据挖掘算法构建霾等级预报分类器，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。
中文关键词: 数据挖掘， KNN， 霾， 预报
Abstract:Based on meteorological data and PM2.5 concentration data of 2013， the statistical analysis shows that temperature, pressure, relative humidity, dew point temperature, surface wind, U wind， V wind and PM2.5 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 frog haze 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.
文章编号： 中图分类号： 文献标志码：
基金项目: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）和首都蓝天行动培育专项（Z141100001014013）共同资助
引用文本：
熊亚军,廖晓农,李梓铭,张小玲,孙兆彬,赵秀娟,赵普生,马小会,蒲维维,2015.KNN数据挖掘算法在北京地区霾等级预报中的应用[J].气象,41(1):98-104.
XIONG Yajun,LIAO Xiaonong,LI Ziming,ZHANG Xiaoling,SUN Zhaobin,ZHAO Xiujuan,ZHAO Pusheng,MA Xiaohui,PU Weiwei,2015.Application of KNN Data Mining Algorithm to Haze Grade Forecasting in Beijing[J].Meteor Mon,41(1):98-104.
熊亚军,廖晓农,李梓铭,张小玲,孙兆彬,赵秀娟,赵普生,马小会,蒲维维,2015.KNN数据挖掘算法在北京地区霾等级预报中的应用[J].气象,41(1):98-104.
XIONG Yajun,LIAO Xiaonong,LI Ziming,ZHANG Xiaoling,SUN Zhaobin,ZHAO Xiujuan,ZHAO Pusheng,MA Xiaohui,PU Weiwei,2015.Application of KNN Data Mining Algorithm to Haze Grade Forecasting in Beijing[J].Meteor Mon,41(1):98-104.