摘要
采用单个时滞非线性神经元结构储备池对不同波形进行识别,研究虚节点个数、反馈强度、输入增益等参数和输入端信噪比对波形识别计算的影响.以方波、正弦波、三角波的识别为任务进行了仿真实验,结果显示:储备池虚拟节点个数为100时最佳;反馈强度和输入增益的参数值分别为0.8和2.2时,储备池的计算性能最佳;在输入端人为加入适度的噪声可降低识别误差,有利于提高波形识别精度.
The single time-delay nonlinear neural reservoir method for waveform recognition is studied. The influence of virtual nodes number, feedback strength, input gain and signal to noise ratio of input layer on waveform recognition is analyzed. Simulation experiments are performed to recognize square waveform, sine waveform and triangle waveform. The results show that when the number of virtual nodes is 100,and the feedback strength and the input gain is 2.1 and 0.8 respectively,computing performance of reservoir is the best ; adding moderate white noise deliberately at input can reduce the recognition error and improve the waveform recognition accuracy.
出处
《内蒙古师范大学学报（自然科学汉文版）》
CAS
北大核心
2016年第6期773-775,780,共4页
Journal of Inner Mongolia Normal University(Natural Science Edition)
基金
内蒙古自治区高等学校科学研究项目(NJZY031)
内蒙古师范大学科研基金项目(2015YBXM004)
关键词
波形识别
储备池
虚拟节点
反馈强度
输入增益
信噪比
waveform recognition
reservoir
virtual nodes
feedback strength
input gain
signal to noise ratio
