A GA-ANN Model for Air Quality Predicting

Abstract

Numerous studies have shown that ANN (Artificial Neural Networks) performs better than traditional regression model on air quality predicting. For better performance, an improved ANN model, called GA-ANN, is proposed, in which GA (genetic algorithm) is used to select a subset of factors from the original set and the GA-selected factors are fed into ANN for modeling and testing. In the experiments, air quality monitoring data and meteorological data (9 candidate factors) of Tianjin, China from 2003 to 2006 are utilized for modeling, and the data in 2007 is utilized for performance evaluation. Three models, including GA-ANN, normal ANN and PCA-ANN, are compared. The correlation coefficients of GA-ANN, which are calculated between monitoring and predicting values are both higher than the other two models for SO2 (sulfur dioxide) and NO2 (nitrogen dioxide) predicting. The results indicate that GA-ANN model performs better than another two models on air quality predicting.