基于支持向量机的细颗粒物日均浓度预测  

Prediction of fine particulate matter average concentration based on support vector machine

作　　者：杨婳妍 吴育联 朱婧巍 易凡茹 吴小涛 YANG Hua-yan;WU Yu-lian;ZHU Jing-wei;YI fan-ru;WU Xiao-tao(College of Mathematics and Statistics,Huanggang Normal University,Huanggang 438000,Hubei,China)

机构地区：[1]黄冈师范学院数学与统计学院,湖北黄冈438000

出　　处：《黄冈师范学院学报》2020年第3期19-25,共7页Journal of Huanggang Normal University

基　　金：黄冈师范学院博士基金项目(201828603);2019年湖北省大学生创新创业训练计划项目(S201910514029)。

摘　　要：为了更加准确的预测大气中的细颗粒物日均浓度,提出了集合经验模态分解(EEMD)-样本熵(SE)和粒子群优化算法(PSO)-最小二乘支持向量机(LSSVM)的组合预测模型。首先采用EEMD方法将原始序列分解成多个相对比较平稳的分量;然后对各分量采用SE进行复杂度分析之后重组,得到若干个新分量;接着,对新分量分别建立采用PSO优化惩罚参数和核参数的LSSVM预测模型,得到各个新分量的预测结果;最后将这些预测结果叠加,得到原始序列的预测值。对上海市的细颗粒物日均浓度采用单一的LSSVM模型和ARIMA模型和本文提出的模型进行预测,并采用多个评价指标进行评价,得出本文提出模型的预测精度更高。In order to predict the daily average concentration of fine particulate matter in the atmosphere more accurately,In this paper,a combination forecasting model based on empirical mode decomposition(EEMD)-sample entropy(SE)and particle swarm optimization(PSO)-least squares support vector machines(LSSVM)is proposed for short-term prediction.Firstly,the EEMD method is used to decompose the original sequence into several relatively stable components;then,these components are recombined after complexity analysis by SE,and several new components are obtained;and then,LSSVM forecasting models with PSO to optimize penalty parameters and kernel parameters are established for each new component,and the forecasting results of each new component are obtained;finally,the predicted values of the original sequence are obtained by superimposing forecasting results of new components.A single LSSVM model and ARIMA model and model proposed in this paper were used to predict the average PM2.5 concentration in Shanghai.Several evaluation indexes were used to evaluate the prediction models.It was concluded that the prediction accuracy of the proposed model in this paper was higher.

关 键 词：细颗粒物日均浓度 集合经验模态分解 样本熵 粒子群优化算法 最小二乘支持向量机 

分 类 号：O24[理学—计算数学]