摘要
针对传统小波神经网络利用随机值作为网络初始参数时,存在网络收敛慢甚至不收敛的问题,该文提出了对网络初始参数进行自相关修正的优化方法,来提高小波神经网络的收敛速度。同时,该文利用了小波函数的降噪特性对原始数据进行预处理,以提高模型的预测精度。将这种改进的小波神经网络模型用于某地铁监测保卫区隧道内的水平位移预测。实验结果表明,改进后的小波神经网络收敛所需迭代次数显著减少,模型的预测精度也更加高。
According to the fact that the traditonal wavelet neural network had the problem of conver- gence when it used random value as network initial parameters, this paper presented the majorization on network initial paremeters to improve the speed of network convergence. And this paper also used wavelet denoising to improve the model＇s prediction accuracy. Then the improved wavelet neural network was used to subway prediction and the experiment＇s results showed that the number of iterations required for con- vergence is significantly reduced and it had more accurate prediction effects.
出处
《测绘科学》
CSCD
北大核心
2017年第9期112-115,共4页
Science of Surveying and Mapping
关键词
小波神经网络
小波降噪
参数优化
变形预测
wavelte neural network
wavelet denoising
majorization of paremeters
deformation prediction
作者简介
陈林（1993-），男，江苏如皋人，硕士研究生，主要研究方向为精密工程测量。E—mail：544911005@qq．com
