基于随机森林和神经网络的空气质量预测

Air Quality Prediction Based on Random Forest and Neural Networks

不同地区的空气质量由于其影响因素不同,通常呈现出不同的变化特征。因污染物的时间长短对预测存在一定的影响,将兰州市空气质量监测中主要的6种污染物(PM2.5、PM10、SO2、NO2、CO、O3)历史数据划分为不同时间的7组特征集,对比分析几种单一模型在7组特征集上对空气质量预测的拟合效果,得到拟合效果最优的单一模型是神经网络模型。考虑到特征选择对空气质量预测的影响,本文建立随机森林组合神经网络模型(RF+MLP)对空气质量指数进行预测,最后将三种模型(RF+MLP、RRFMLP、MLP)对空气质量指数预测的拟合结果进行了比较分析,结果表明,RF+MLP模型对兰州市空气质量指数的预测拟合度在整体上是最好的。

Due to different influencing factors,the air quality in different regions usually exhibits different characteristics of change.Because the duration of pollutants has a certain impact on the prediction,the historical data of six major pollutants(PM2.5,PM10,SO2,NO2,CO,O3)in air quality monitoring in Lanzhou City are divided into seven sets of feature sets at different times,and the fitting effect of several single models on air quality prediction on seven sets of feature sets is compared and analyzed.It is found that the single model with the best fitting effect is the neural network model.Considering the impact of feature selection on air quality prediction,this paper establishes a random forest combination neural network model(RF+MLP)to predict the air quality index.Finally,the fitting results of the three models(RF+MLP,RRFMLP,MLP)for air quality index prediction were compared and analyzed.The results showed that the RF+MLP model had the best overall fitting degree for predicting the air quality index in Lanzhou City.