An Experimental Study of Haze Prediction Method in Nanjing

The Support Vector Machine (SVM) is a new machine learning method based on the statistical learning theory and it is very useful to solve nonlinear problems of short time series. Based on the daily observations obtained from the Nanjing meteorological station and the daily measured contamination data obtained from the Nanjing environmental quality monitoring station from 2004 to 2007，prediction models of hazeday classification and visibility at 1400 LST in haze days in Nanjing are built by the SVM method. The results show that the threat scores（Ts）of hazeday classification forecast are all over 04 and the precision of visibility at 1400 LST in haze days can reach 86% by considering 3 km as the error bounds. Otherwise，the forecast models which are revised by the new data at 0800 LST of that day are better than the beginning forecast models according to the results. Both SVM forecast models perform satisfactorily and can refer references to real business.