文档介绍：基于递归神经网络的LS-SVM硬件实现
与实验研究*
刘涵,叶平
(西安理工大学自动化与信息工程学院西安 710048)
摘要:在标准支持向量机(SVM)学习神经网络的基础上,将最小二乘支持向量机(LS-SVM)与递归神经网络相结合, 提出一种新的最小二乘支持向量机学习神经网络。该网络直接采用Lagrange乘子进行训练,消除了标准SVM神经网络中的线性部分,可用于进行分类和回归学习。并且其拓扑结构更适合于用简单的硬件模拟电路实现。对两种网络的稳定性进行了证明,并设计了相应的硬件电路,最后通过Simulink、Pspice仿真和硬件电路实验证明了所提出的方法是有效的。
关键词:最小二乘支持向量机;递归神经网络;模拟电路
中图分类号:TP183 文献标识码:A 国家标准学科分类代码:
Hardware implementation and experiment research of least square support
vector machine based on recurrent work
Liu Han, Ye Ping
(School of Automation and Information Engineering, Xi’an University of Technology, Xi’an 710048, China)
Abstract:A new work for least squares support vector machines (LS-SVM) bines LS-SVM with recurrent work is presented, based on standard SVM work. work is trained using Lagrange multiplier directly, which can eliminate the nonlinear parts of the work of standard SVM. The work can be used for classification and regression learning, whose topology adapts to analog circuit implementation easily. The stabilities of two kinds of works are also proven and corresponding analog circuits are proposed in the paper. The results of Simulink and Pspice simulations and hardware circuit experiments all illustrate the effectiveness of the proposed works.
Key words:LS-SVM; recurrent work; analog circuit
1 引言
支持向量机(support vector machines, SVM)采用结构风险最小化的思想和方法,具有良好的推广能力、极低的分类和逼近误差、数学上的易处理性和简洁的几何解释等优点,已被广泛作为一种