Research On Prediction Of PM₁₀ Pollution Based On RBF Neural Network

The prediction of PM₁₀ pollution has been studied according to the PM₁₀ concentration data daily and the ground meteorological data daily that monitored from June 1 of 2001 to March 15 of 2006.Analyed the PM₁₀ concentration's change anualy, monthly, heating period and non-heating period, it was found that PM₁₀ concentration in heating period was more greater than in non-heating period. It also analyed reasons of PM₁₀ pollution.PM₁₀ pollution is closely-related with meteorological factors, through correlation analysis between PM₁₀ pollution and meteorological factors, PM₁₀ concention prediction model can be established. But the relationship between PM₁₀ pollution and meteorological factors are ofen nonlinear, so it is difficult to establish a precise mathematical model. Traditional mathematical model and statical method's result are not well. If it want to predict PM₁₀ concention more precise, a method that can catch nonlinear varation's displine must be used. Because RBF neural network is a powerful tool to deal with nonlinear problem, so it is used for establishing Xi'an city's PM₁₀ concentration prediction model in heating period and non-heating period. The correlation between PM₁₀ concentration and meteorological factors were analyed by RBF neural network before building the RBF neural network model. It was established two kinds of PM₁₀ pollution models by using RBF neural network's Orthogonal Least Mean Squares algorithm and Nearest Neighbor-clustering algorithm. Tested the model's performance, the average precision of heating period's Orthogonal Least Mean Squares model was 64.09%, Nearest Neighbor-clustering model was 74.47%, the average precision of non-heating period's Orthogonal Least Mean Squares model was 74.01%, Nearest Neighbor-clustering model was 79.48%.