摘要
为了解安徽省PM2.5分布特征、定量分析机器学习算法在预测PM2.5浓度方面的准确性.针对安徽省78个空气及气象监测站点的数据进行统计分析,并从时间因子、其它空气污染物浓度、气象因子三个方面,筛选出月、周、时,PM10、CO、SO2、N20,风速、气温、相对湿度、降水量、当前PM2.5浓度,共计12个自变量,利用K折交叉验证,分别构建基于支持向量机、神经网络的预测模型.表明安徽省PM2.5浓度总体呈现出北高南低,冬高夏低的特征,浓度值较高的月份出现在1、2、11、12,周变化规律各地有异同,时变化规律各地较为一致,8时至17时下降,其余时段上升.模拟结果显示支持向量机模型预测效果较好,其预测值与实测值的均方根误差控制在20以内,拟合指数在0.8以上.
To understand the distribution characteristics of PM2.5 in Anhui Province and to quantitatively analyze the accuracy of machine learning algorithms in predicting PM2.5 concentration.Statistical analysis was carried out on the data of 78 air and meteorological monitoring stations in Anhui Province,and the monthly,weekly and time,PM10,CO,SO2 were selected from three aspects:time factor,other air pollutant concentration and meteorological factors.N2 O,wind speed,temperature,relative humidity,precipitation,current PM2.5concentration,a total of 12 independent variables,respectively,based on support vector machine,neural network prediction model.It indicated that the concentration of PM2.5 in Anhui Province showed the characteristics of high north and low south,low in winter and high in summer,and the month with high concentration appeared in 1,2,11,and 12.The regularities were more consistent everywhere,falling from 8:00 to 16:00 and rising during the rest of the period.The simulation results show that the support vector machine model has better prediction effect within 15 hours,and the root mean square error between the predicted value and the measured value is controlled within 20,and the fitting index is above 0.8.
作者
杨小兵
杨峻
华华
黄晓英
张成扬
YANG Xiao-bing;YANG Jun;HUA Hua;HUANG Xiao-ying;ZHANG Cheng-yang(Xuancheng Meteorology,Anhui Meteorology,Xuancheng 242000,China;School of Automation,Southeast University,Nanjing 210000,China;Guangxi Climate Center,Guangxi Meteorology,Nanning 530000,China)
出处
《数学的实践与认识》
北大核心
2020年第1期285-291,共7页
Mathematics in Practice and Theory
基金
安徽省气象局硕博士工作启动经费项目“CIMISS结合Spark在业务中的应用研究”(RC201620).
关键词
PM2.5
时空分布
支持向量机
神经网络
预测
PM2.5
space-time distribution
support vector machine
neural network
prediction
