城市交通道路氮氧化物浓度的CART回归树预测研究

提出了基于CART回归树的氮氧化物(NO_x)浓度预测模型,利用杭州市延安路路边空气质量监测站2016年6—9月空气污染物监测数据和同期延安路路段车辆抓拍识别数据,通过数据处理、影响因素分析及CART回归树构造,搭建了NO_x浓度预测模型.实验分析结果表明,相对于支持向量机和BP神经网络预测模型,基于CART回归树的NO_x浓度预测模型的预测精度有大幅度提升,可决系数在0.92以上;同时,对环境条件差异较大的G20会议期间NO_x浓度进行预测分析,结果表明,CART回归树方法的预测精度比其它方法更高,能够适应不同条件下的预测需求.

A CART-based approach to predict nitrogen oxide concentration along urban traffic roads

A new prediction model for predict nitrogen oxide (NO_x) concentration along urban traffic roads was proposed based on the classification and regression tree (CART). The sample data during June to September, 2016 in Hangzhou, was collected from atmospheric monitoring station and traffic electronic police system. The proposed model is established to predict NO_x concentration by data processing, analysis of influence factors and construction of CART algorithm. We compare and analyze the results using the proposed CART-based approach, a support vector regression (SVR) based prediction model and a back propagation (BP) neural network based prediction model. The outcome shows the CART-based approach can make the most accurate predictions of NO_x concentration and its determination coefficient is more than 0.92. It also performs the best when the enviromental circumstance is sharp changed from the usual, such as the period of the G20 Hangzhou Summit. The results suggest the CART-based approach can adapt to the various conditions with higher forecast accuracy of NO_x concentration than other methods.