A Haze Prediction Algorithm Based on PCA-BP Neural Network

In this paper, we proposed a novel haze forecast model, which combine the principal component analysis with back-propagation neural network to solve the air quality problem in China. Comprehensive variables are obtained by dimension reduction on many predictive factors through principal component analysis (PCA). Then we use the comprehensive variables as the input of back-propagation (BP) neural network. Through this process, the correlation among the original predictors can be eliminated and the structure of neural network can be simplified. The simulation results show that the average prediction error of the prediction model by using principal component analysis combined with BP neural network is less than 10%, far lower than the results of the traditional prediction method which only use a single index, PM2.5 daily average concentration, to judge whether it is haze day.