Study on Air Fine Particles Pollution Prediction of Main Traffic Route Using Artificial Neural Network

Abstract

In this paper the feasibility of artificial neural network technology for air fine particles pollution prediction of main traffic route was discussed. The concentration data of PM2.5, PM5 and PM10 were measured in Zhongshan road, the main traffic route of Chongqing, China. Parameter Φ of emission capacity of motor vehicles was used as the independent variable of prediction model. RBF and BP neural network were used to simulate the concentration of fine particles of different sizes. The results show that: (1) Prediction results of PM of different sizes are different, the simulating data of PM2.5 using RBF networks are better than those of PM5 and PM10; (2) The simulation effect of RBF neural network is related to maximum nerve cell number of network and the distribution density of radial basis function. When the maximum nerve cell number is 13 and the distribution density of radial basis function is 0.9, the simulation result of PM2.5 is best; (3) Using three hidden layers and Levenberg-Marquardt calculation method of BP neural network, good simulation effect could be achieved; (4) For PM2.5, the correlation coefficient between simulating data of testing sample and testing data are 0.94 and 0.91, the ratio of training error and testing error are 0.75 and 1.59 each by RBF and BP neural network. All above show that PM2.5 of main traffic route come mainly from vehicle emission. The two neural network established herein can be used to predict pollution of PM2.5.