基于RF-Kmeans-LIBSVM的乌鲁木齐市颗粒物浓度预测研究  

Particle concentration forecast of Urumqi based on RF-Kmeans-LIBSVM

作　　者：李爱英 LI Aiying(Xinjiang Uygur Autonomous Region Environmental Engineering Assessment Center,Urumqi 830016,China)

机构地区：[1]新疆维吾尔自治区环境工程评估中心,新疆乌鲁木齐830016

出　　处：《环境保护科学》2022年第4期118-124,共7页Environmental Protection Science

基　　金：甘肃省自然科学基金资助项目(21JR7RA501)。

摘　　要：为了准确预测空气中颗粒物的浓度变化情况,减少空气污染给居民的生产生活带来的危害,该研究提出一种基于RF-Kmeans-LIBSVM的大气颗粒物浓度预测模型。首先采用RF算法对影响PM_(2.5)和PM_(10)浓度的因子进行重要性评估,选择出影响最大的2个因子作为聚类属性,然后采用Kmeans算法对空气污染监测数据进行聚类,把PM_(2.5)和PM_(10)序列划分为相似性较高的若干类,最后运用经聚类分析之后的训练样本建立PM_(2.5)和PM_(10)浓度预测模型。以乌鲁木齐市监测点2015年1月1日~2020年12月31日的PM_(2.5)和PM_(10)浓度日均监测数据为例,使用改进方法和传统方法分别进行预测。结果表明:与传统支持向量机相比,改进后的模型的预测准确率明显提升,对于PM_(2.5),误差评价指标MAE和RMSE分别下降33.1%和26.5%;对于PM_(10),误差评价指标MAE和RMSE分别下降15.7%和12.7%。研究说明利用RF-Kmeans聚类分析的方法来提高传统支持向量机在PM_(2.5)和PM_(10)浓度预测中的泛化能力具有可行性。In order to accurately predict the concentration of particulate matter in the air and reduce the harm caused by air pollution to the residents,this study proposes an atmospheric particulate matter concentration prediction model based on RF-Kmeans-LIBSVM.First,the RF algorithm is used to evaluate the importance of the factors that affect PM_(2.5),PM_(10)concentration.The two serious influential factors in the factor set are selected as clustering attributes,and then the Kmeans algorithm is used to cluster the air pollution monitoring data.PM_(2.5),PM_(10)sequences are divided into several categories with the high similarity.Finally,the training samples after cluster analysis are used to establish a PM_(2.5),PM_(10)concentration prediction model.Taking the daily average PM_(2.5),PM_(10)concentration monitoring data from the monitoring point in Urumqi from January 1,2015 to December 31,2020 as an example,the forecast is performed by the improved method and the traditional method.The results show that compared with the traditional support vector machine,the prediction accuracy of the improved model is significantly increased.For PM_(2.5),the error evaluation indexes MAE and RMSE decrease by 33.1%and 26.5%,respectively.For PM_(10),the error evaluation indexes MAE and RMSE decrease by 15.7%and 12.7%,respectively.The study shows that it is feasible to use the RF-Kmeans cluster analysis method to improve the generalization ability of traditional support vector machines in PM_(2.5),PM_(10)concentration prediction.

关 键 词：PM_(2.5) PM_(10) 聚类分析 支持向量机 预报 

分 类 号：P456[天文地球—大气科学及气象学]