Rapid Prediction of Atmospheric Pollutants in Local Region Through Optimized RBF Methods

Addressing the problem that multiple complex factors affect the emission and accumulation of Main Air Pollutants in local region, the optimized RBF (Radial Basis Function) neural networks methods are suggesting to find local atmospheric pollutants concentration and its relationship with real-time tracking in this paper. Firstly it acquires meteorological conditions, such as complex ground conditions and multiple- emission sources at the scale of local areas and residential areas; secondly RBF connection mechanism is reconstructed to effectively utilize the linear information between input and output neurons, improve the generalization performance of the network, and quickly construct high-precision self-organizing RBF neural networks; thirdly an efficient and accurate pollutant concentration evaluation model is constructed to evaluate the concentration of pollutants in local region using emission data predicted by energy economic models. Experimental results indicate that our RBF mechanism is efficient for it can not only maintain the stability of the network output with a compact structure in the entire training process where no global knowledge is required to be maintained, but also has superior generalization ability and faster convergence speed which are bounded by the number of neighbors with respect to the total time for satisfying online prediction.