| 基于混合优化的RBF神经网络集成的降水预报模型 |
| |
| 引用本文： | 蒋林利.基于混合优化的RBF神经网络集成的降水预报模型[J].柳州师专学报,2012(2):113-119. |
| |
| 作者姓名： | 蒋林利 |
| |
| 作者单位： | 柳州师范高等专科学校 数学与计算机科学系, 广西 柳州 545004 |
| |
| 基金项目： | 柳州师专科研基金资助项目“流行学习及降维方法的研究”(LSZ2011B007) |
| |
| 摘 要： | 针对传统的单个RBF神经网络集成中个体的隐节点个数和初始参数难以客观确定的不足,为了提高泛化能力,提出一种以高斯核函数的混合优化的RBF神经网络的方法,首先引入正交最小二乘法动态客观的获取数据中心的个数、数据中心及权值;然后通过计算隐层中心点间最小距离作为扩展常数;最后使用剃度法调节权值、中心及扩展常数使网络参数和结构达到最优.该方法结合了正交最小二乘法和剃度算法的优点,通过从结构和算法两方面的调整提升了单个的传统的RBF网络的性能.并将上述优化混合的RBF神经网络与主成分分析方法相结合建立模型.本文以广西5月逐日降水事先初选的众多预报因子进行主成分分析算法提取有效的几个综合因子,然后使用混合算法优化的径向基网络建立降水预测模型.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,具有一定的普遍适用性.
|
| 关 键 词： | 主成分分析 混合优化的RBF神经网络 核函数 预测 |
On Precipitation Forecast Model Based on Hybrid Optimized RBF Neural Network Integration |
| |
| JIANG Linli.On Precipitation Forecast Model Based on Hybrid Optimized RBF Neural Network Integration[J].Journal of Liuzhou Teachers College,2012(2):113-119. |
| |
| Authors: | JIANG Linli |
| |
| Institution: | JIANG Linli(Department of Mathematics and Computer Science,Liuzhou Teachers College,Liuzhou,Guangxi,545004 China) |
| |
| Abstract: | This paper,aiming at the deficiencies of being difficult to objectively determine the traditional single RBF neural network integration of the individual in the number of hidden nodes and initial parameters,puts forward a kind optimization of RBF neural network method which is on Gauss kernel function to improve the generalization ability.It firstly introduces the orthogonal least square method dynamic objective data acquisition center number,data center and weight;then computes the minimum distance between hidden layer center,which is considered as an extension of constant;finally uses gradient method to adjust weights,centers and extended constant in order to make the network parameters and structure optimal.The method,combining the advantages of the orthogonal least squares method and gradient algorithm,promotes the performance of the single traditional RBF network from the adjustment of the two aspects of structure and algorithm and establishes model by combining the two methods of the optimization of mixed RBF neural network and principal component analysis.It primarily chooses numerous forecast factors from the daily precipitation of Guangxi in May and principal component analysis algorithm for extracting effective several comprehensive factors,then builds model for rainfall forecast by using a hybrid optimization algorithm of RBF neural network.The result shows that,the model has a good convergence and generalization ability and a certain universal applicability in the prediction performance which is obviously superior to the earlier T213 precipitation forecast. |
| |
| Keywords: | principal component analysis hybrid optimization RBF neural network kernel function forecasting |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
