基于机器学习的PM2.5短期浓度动态预报模型

发布时间：2018-03-17 00:37

本文选题：机器学习　切入点：粒子群优化算法　出处：《计算机应用》2017年11期 　论文类型：期刊论文

【摘要】：针对目前现有的PM2.5模式预报系统的预报值偏离实际浓度较大的问题,从上海市浦东气象局获得2015年2月至7月的PM2.5实况观测浓度、PM2.5模式预报(WRF-Chem)浓度和5个主要气象因子的模式预报数据资料,联合应用支持向量机(SVM)和粒子群优化(PSO)算法建立滚动预报模型,对PM2.5未来24小时浓度进行预报,同时对未来一天的昼、夜均值及日均值浓度进行预报,并与径向基函数神经网络(RBFNN)、多元线性回归法(MLR)、模式预报(WRF-Chem)作对比。实验结果表明,相比其他预报方法,所提出的SVM模型较大提高了PM2.5未来1小时浓度预报精度,这与此前的研究结论相符;所提模型能对PM2.5未来24小时浓度进行较好的预报,能对未来一天的昼均值、夜均值及日均值进行有效预报,并且对未来12小时的逐时浓度及未来一天的夜均值浓度的预报准确度较高。
[Abstract]:In view of the problem that the prediction value of the existing PM2.5 model forecasting system deviates from the actual concentration greatly, From February 2015 to July, the data of PM2.5 observed concentration and PM2.5 model forecast of WRF-Chemand five main meteorological factors were obtained from Shanghai Pudong Meteorological Bureau. A rolling prediction model is established by using support vector machine (SVM) and particle swarm optimization (PSO) algorithm to predict the concentration of PM2.5 in the next 24 hours, and to predict the concentration of day, night and day in the next day. Compared with the radial basis function neural network (RBFN), the multivariate linear regression (MLR) and the model prediction (WRF-Chem), the experimental results show that the proposed SVM model can greatly improve the accuracy of PM2.5 concentration prediction in the next 1 hour compared with other prediction methods. This is consistent with previous studies. The proposed model can predict the concentration of PM2.5 in the next 24 hours, and can effectively predict the day average, night average and daily average value of the next day. The prediction accuracy of hourly concentration in the next 12 hours and the night-mean concentration in the next day is higher.
【作者单位】：浙江师范大学数理与信息工程学院;上海市气象局中心气象台;
【基金】：国家自然科学基金资助项目(41575046) 浙江省科技厅公益性技术应用研究计划项目(2016C33010) 浙江省金华市科技计划项目(2014-3-028)~~
【分类号】：X513

【相似文献】