Papers

Full Papers

1. Pollution analysis and prediction based on Voronoi Maps and neural networks

by Zejia Zheng,Liwen Zhou,Longfei Ren (Fudan University)

Abstract: Attention of this paper focuses on the evaluation of pollution condition of a city and finding the pollution sources using data gathered from observation stations all over the city. We assume every observation data represents the condition of pollution in a certain area, which was given by formulating a Voronoi map of the region with the center points of each Voronoi area representing the observation station. By calculating the area of each Voronoi area, we assigned weight to each observation station, thus modifying the traditional method to evaluate the pollution condition. We further assume that pollutants will only be transferred between adjacent Voronoi areas. So the existence of pollution transfer can be represented by drawing a line between adjacent observation stations, thus giving us the Delaunay Triangulation of the city. We substitute the observation station as neurons, the edge of the Delaunay Triangulation as synapses, thus getting a neural network. According to our assumption, the transfer rate of pollution between two neurons, or the weight of the neural network, depends on the difference of their density of pollution, altitude, and the distance between the two observation stations. We show that prediction and backtrack pollutant transfer can be realized naturally from our network.

Index Terms: Neural Networks, Voronoi Map, Pollutant

Page: 1-6

Full Text: PDF