Author(s)

Yuxiang Wang

Corresponding Author

Yuxiang Wang

ABSTRACT

Recently, air quality has always been the focus of attention in China, and the air quality has gradually improved under constant efforts. This paper aims to establish a reasonable model to predict and analyze PM2.5 concentration data in Chengdu. After an analysis of the main influencing factors of PM2.5 in Chengdu urban area of recent years, 39 sample data from 8 meteorological observation points in Chengdu in February 2019 are selected, factors including PM2.5 concentration, humidity and wind speed as input and output. The BP neural network model is used to predict the concentration of PM2.5 in Chengdu. Due to the large dimension of the input elements, principal component analysis is introduced in the input layer to achieve the goal of dimensionality reduction. The principal components of the principal component analysis method is then used to obtain the optimal initialization weights and thresholds by genetic algorithm, and the activation function uses Leaky-ReLU. By training with sample data, the network model structure is determined, and prediction accuracy is tested by using the sample data. The results show that the optimized BP neural network model has a good performance in PM2.5 concentration prediction with high precision.

KEYWORDS

PM25, BP Neural Network, PCA, GA, Leaky-ReLU