论文部分内容阅读
为了实现超低频振动速度测量,提出补偿其幅频特性的小波神经网络方法.该方法以振动速度传感器动态实验数据为基础,通过小波神经网络训练来确定传感器幅频特性补偿网络.介绍振动速度传感器幅频特性补偿原理,分析网络的拓扑结构,给出网络参数训练和初始化方法.采用引入动量项的最速下降法训练网络权值、尺度因子和平移因子,将小波网络参数的初始化与小波类型、小波时频参数和学习样本等联系起来.结果表明,采用小波神经网络进行振动速度传感器幅频特性补偿具有良好的鲁棒性,并能实现在线补偿,网络训练的速度和精度优于同等规模的BP网络,在测试领域有重要的实用价值.
In order to realize the ultra-low frequency vibration velocity measurement, a wavelet neural network method is proposed to compensate the amplitude-frequency characteristics of the sensor. The method is based on the dynamic experimental data of the vibration velocity sensor to determine the amplitude-frequency characteristic compensation network of the sensor through wavelet neural network training. Frequency characteristics of the compensation principle, the network topology analysis, network parameters training and initialization method is given.With the steepest descent method introduced the momentum term training network weight, scale factor and the translation factor, the wavelet network initialization and wavelet parameters, Wavelet time-frequency parameters and learning samples, etc. The results show that using wavelet neural network to compensate the amplitude-frequency characteristics of the vibration velocity sensor has good robustness and online compensation, the speed and accuracy of the network training is better than that of the same scale BP network, in the test area has important practical value.