摘要
针对BP小波神经网络模型易陷入局部极小和收敛速度慢等问题,结合Morlet小波函数、训练样本数量进行权值和阈值设置,引进隐含层饱和度并构建新的误差函数,以此提高模型收敛速度和预测效果。从模型精度、后验差比值和训练次数这三个指标进行对比分析。结果表明,改进的模型预测效果更满意。
Aiming at the problem often faced with by BP Wavelet neural network model that local is minimum and rate of convergence is slow and combining with that Morlet Wavelet function and training sample number are set weight and threshold value,hidden layer saturability is used to build new error function so that the model rate of convergence and predicting effect can be improved.By comparing model precision,a posterior difference ratio and training frequency,the result shows that the result of improved model is satisfactory.
作者
余兰
岳建平
Yu Lan Yue Jianping(School of Earth Sciences and Engineering, Hohai University, Nanjing 211100 ,China)
出处
《甘肃科学学报》
2016年第5期38-41,共4页
Journal of Gansu Sciences
关键词
沉陷预测
小波神经网络
模型精度
后验差比值
训练次数
Sink prediction
Wavelet neural network
Model precision
A posterior difference ratio
Training frequency