基于动态基函数的强回归小波神经网络  

Strong Regression Wavelet Neural Network Based on the Dynamic Basis Functions

作　　者：李翔[1] 朱全银[1] 

机构地区：[1]淮阴工学院计算机工程学院,江苏淮安223003

出　　处：《北京工业大学学报》2014年第7期1084-1090,共7页Journal of Beijing University of Technology

基　　金：国家星火计划资助项目(2011GA690190);淮安市科技支撑计划资助项目(HASZ2013008)

摘　　要：针对小波神经网络(wavelet neural network,WNN)难以选取合适小波基函数和确定隐含层节点数等问题,提出使用集成学习改进小波神经网络的方法,提高小波神经网络容错能力和自学习能力.本方法首先通过降维、归一化预处理样本数据并确定测试数据分布权值;然后通过随机选取不同的小波基函数构造出异构小波神经网络序列并反复训练样本数据;最后使用AdaBoost算法集成学习生成强回归小波预测器.对UCI数据库中数据集进行仿真验证,实验结果表明:本方法比传统小波神经网络预测平均误差减少30%以上,有效地提高了小波神经网络的预测精度和泛化能力.In view of the wavelet neural network （WNN） that is difficult to select the appropriate wavelet functions and determine the hidden layer nodes and other issues, a method of using ensemble learning with WNN was put forward to improve the fault-tolerant ability and self-learning ability. First, the method performed the sample data using the dimensionality reduction and normalization method, and determined the distribution weights of test data. Second, it randomly selected different wavelet basis functions to construct heterogeneous predictors of WNN and repeatedly trained the sample data. Finally, AdaBoost algorithm ensemble learning is used to form a new strong predictor. A simulation verification for the database of UCI was carried out. Results show that this method reduceds the average error value by more than 30% compared with the traditional wavelet neural network, and improves the forecasting accuracy and generalization ability of WNN.

关 键 词：小波神经网络 强预测器 集成学习 ADABOOST算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]