Study On The Road Network Selection Methods Based On The BP Neural Network

BP neural network is utilized to solve the problem of too many vague measures to improve the effect of the selection of the road networks, which was prepared for the following parallelization.The constructional algorithm of stroke was researched to describe the network hierarchy structure of road networks better.This article is the important research content of the national natural science foundation programs(No. 41071288),named "Study on parallel computational methods of cartography generalization based on network hierarchy structure".Cartographic generalization was a challenging and creative research in the field of cartography and GIS.No breakthrough has been made in cartographic generalization,which was because that a large number of problems in cartographic generalization are vague in nature and can’t be described by certain mathematical models or formulas.Intelligent approachs are the fundamental ways of the research of cartographic generalization.The key of intelligent cartographic generalization is the formal expression and acquisition of knowledge,which is difficult now.There are many artificial intelligent techniques for cartographic knowledge acquisition.As an important artificial intelligence technology,artificial neural network has great prospects of application in cartographic generalization.BP neural network was utilized for the research of selection of road network,and the selective results of multi-scale road networks were used as the training sample of the BP neural network.The prevenient research achievements and methods of the road network’s selection were studied and reviewed,then stroke was utilized as the basic selection unit.Considering the geometric,semantic and topological properties of the road networks,length,degree centrality,betweenness centrality and closeness centrality were selected as the selective indicators of road network and the inputs of the BP neural network.A suitable structure of BP neural network to estimate the importance of the roads was obtained.Finally,based on the importance of strokes,the strokes were selected by the "selection principle" according the quantity.The experimental data of this article was from road networks of the national atlas of USA,including two scales-1:2,000,000 and 1:1,000,000,and the 1:2,000,000 one was generalized through the 1,000,000 one.Some of the roads in the road network of Ohio were selected as the training sample.lt was found that a BP neural network with one hidden layer and eight neurons has the best effect through training,which is available for the estimation of stroke importance.Experiments were made respectively according to the type of road networks like grid type, radial type and free type,finally a whole experiment was made by the road network of Illinois.It was proved to maintain a better overall structure,better connectivity and better road network density distribution.This method turn out to solve the problem of too many vague measures in selection of road network effectively,and the method that the training samples come from multi-scale map data is advisible,which can be used by other researchers.