摘要
多层前馈网络是目前研究得最多和应用最广泛的神经网络之一，其基本算法为误差反向传播（ＥＢＰ）算法，但存在收敛速度慢和局部极小的问题。本文利用递归最小二乘算法来训练多层前馈网络，ＲＬＳ算法具有收敛速度快，抗噪声能力强等优点，还克服了常规ＢＰ算法中学习率选取困难的缺点。仿真结果说明了本文方法的有效性。
Multi-layered feedforward network, with errorback-propagation as its basic algorithm,is one of themost researched and widely used neural networks inrecent years; but its such a kind of algorithm still hasthe shortage of slow convergence speed and the problem of local minimization.In this paper,we train thenetwork by using recursive least squares algorithm,which has a fast parameter convergence speed and agood robustness agaist noise and disturbance, andwhich can also eliminate the need for an arbitary stepsize. A simulation is also given to show the effectiveness of this method.
出处
《自动化与仪器仪表》
1996年第6期5-7,共3页
Automation & Instrumentation
基金
高等学校博士点基金
