基于二分搜索密度峰算法的RBFNN在月降水预报中的应用 后印本

Abstract: Aiming at the shortages of the RBFNN network structure and initial data center being difficult to objectively determined, this paper puts forward using two searching density peak clustering algorithm (TSDPCA) to find data center value and number as the initial parameters of RBFNN and the number of hidden layer nodes quickly, and finally using gradient descent method of optimization the structure of RBFNN and various parameters to set up precipitation forecast model, which was applied in forecasting monthly rainfall of Guagnxi, aiming to testing the effectiveness of the model. The result shows that the mean relative error of TSDPCA-RBFNN forecasting was decreased by 10%-35% compared with K-RBFNN and OLS-RBFNN model, which has better prediction performance.