论文部分内容阅读
针对传统小波神经网络(wavelet neural network,WNN)受隐含层节点数影响大、网络误差易陷入局部极小、预测结果不稳定的问题,提出使用GentleAdaBoost和小波神经网络相结合的方法,提高网络预测精度和泛化能力。该方法首先对样本数据进行预处理并初始化测试数据分布权值;然后通过选取不同的隐含层节点数、小波基函数构造出不同类型的小波神经网络弱预测器序列并对样本数据进行反复训练;最后使用GentleAdaBoost算法将得到的多个小波神经网络弱预测器组成新的强预测器并进行回归预测。对UCI数据库中数据集进行仿真实验,结果表明,本方法比传统小波神经网络预测平均误差减少40%以上,有效地提高了神经网络预测精度,为小波神经网络应用提供借鉴。
Aiming at the problem that the traditional wavelet neural network (WNN) is greatly affected by the number of hidden layer nodes, the network error tends to fall into a local minimum and the prediction result is not stable, a combination of GentleAdaBoost and wavelet neural networks is proposed to improve the network Predictive accuracy and generalization ability. The method firstly preprocesses the sample data and initializes the distribution weight of the test data. Then, by selecting different hidden layer nodes and wavelet basis functions, we construct different types of weak predictor sequences of wavelet neural network and repeatedly train the sample data Finally, we use GentleAdaBoost algorithm to form a new strong predictor by using the multiple wavelet neural network weak predictors and make regression prediction. The simulation experiments on the datasets in UCI database show that the proposed method can reduce the average prediction error by more than 40% compared with the traditional wavelet neural network, effectively improve the prediction accuracy of the neural network, and provide a reference for the application of the wavelet neural network.