基于PCA-LLE组合的优化RBF神经网络降水预测模型  

On Precipitation Forecast Model Based on PCA-LLE algorithm and Optimized RBF Neural Network

作　　者：罗芳琼[1] 吴春梅[1] 

机构地区：[1]柳州师范高等专科学校数学与计算机科学系,广西柳州545004

出　　处：《柳州师专学报》2013年第2期126-130,共5页Journal of Liuzhou Teachers College

基　　金：广西教育厅科研基金(201204LX506);柳州师范高等专科学校科研基金(LSZ2011B007)

摘　　要：提出了主成分分析(PCA)和局部线性嵌入降维方法(LLE)组合的优化RBF神经网络降水预测模型,首先利用主成分分析(PCA)和局部线性嵌入降维方法(LLE)组合对众多气象物理因子降维提取有效因子,再将这些综合有效因子组成的特征空间矩阵作为优化的RBF神经网络的输入矩阵,从而建立网络模型.以此对广西5月三个不同区域平均日降水量进行预报实验,结果表明,该模型具有较好的收敛效果和泛化能力,在预报性能上明显优于同期的T213降水预报、PCA预报、LLE预报,预报稳定性好,预报准确率较高,具有一定的普遍适用性.A precipitation forecast model based on PCA - LLE algorithm and optimized RBF neural network is proposed in the paper. The principal component analysis and the locally linear embedding （LLE） algorithm are presented to extract the precipitation feature, and then the factor matrix with precipitation feature of each sample is trained in the optimized RBF neural networks. For further illustration, the model is employed to test the rainfall data of three different regions in Guangxi in May. The result shows that the model has a good conver- gence and generalization ability and a certain universal applicability in the prediction performance, which is obviously superior to the earlier T213 precipitation forecast and PCA precipitation forecast and LLE precipitation forecast. It is superior to those obtained by other models presented in this study in terms of the same evaluation measurements.

关 键 词：部线性嵌入 主成分分析 径向基神经网络 降维 预测 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]