基于核偏最小二乘回归神经网络集成降水预测模型  

Back Propagation Integrated with Kernel Partial Least Square for Raining Forecast Model

作　　者：陆克盛[1] 汪灵枝[2] 

机构地区：[1]广西民族师范学院数学与计算机科学系,广西崇左532200 [2]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：《沈阳农业大学学报》2013年第3期345-348,共4页Journal of Shenyang Agricultural University

基　　金：国家自然科学基金项目(11161029);广西省高校一般资助项目(201203YB186;200103YB147;2013YB262)

摘　　要：为了提高降水预测的精确度和稳定性,提出一种新颖的基于核偏最小二乘回归的径向基神经网络集成降水预测模型。该模型通过Bagging技术和Boosting技术把原始数据集分成不同的训练数据集,并利用该训练数据集和不同核函数的径向基神经网络进行预测处理,再将核偏最小二乘回归对不同的训练结果进行集成。研究结果表明:核偏最小二乘回归集成模型有效提高神经网络集成的泛化能力,预测精度高,稳定性好,具有应用推广前景。In order to improve forecasting application precision, this research presents a novel prediction model based on kernel partial least-squares and basis function of neural network. The model through Bagging and Boosting technology to divide the original training data set into different sets of data, and uses the different set of training data and radial basis function neural network for prediction of different kernel functions, then kernel partial least-squares regression to integrate different training results. Research results showed that integrated kernel partial least-squares regression model effectively improved the generalization ability of neural network ensembles, and forecast application high precision, good stability and promotion prospects.

关 键 词：核偏最小二乘回归 神经网络集成 降水预测 

分 类 号：TM932[电气工程—电力电子与电力传动]