Study On SOM Neural Network And Its Application In Hydrologic Regionalization

In this paper, principal component analysis, SOM neural network and SOM neural network based on agglomerative hierarchical clustering were studied. The basic data obtained from hydrologic network of Jiangxi province were used for calculation.According to principal component analysis, the contribution rate of every principal component was got, and the contribution rate was introduced into SOM neural network. By weighted Euclidean distance, winner of neurons was found during competitive processing. Weighted SOM neural network was proposed, filling the gap of neglecting the difference of every characteristic component essentiality in traditional SOM neural network.Taking the data of hydrologic network of Jiangxi for instance, comparison between weighted SOM neural network and traditional SOM neural network was made, Conclusions were as follows:1) During competitive processing, weighted Euclidean distance was advisable;2) Weighted SOM neural network was better in clustering accuracy;3) The clustering result was sensitive to initial learning rate.Thanks to the hierarchical structure of hierarchical clustering, the conclusion of extending SOM neural network in hydrologic regionalization is better for moving of hydrologic data.