摘要
首先介绍支持向量机和神经网络方法及其在内部网络训练上的不同.分别利用支持向量机和神经网络对高斯粗糙面的均方根高度和相关长度进行反演.通过仿真结果和误差对比分析,发现在小样本情况下,支持向量机的反演结果比神经网络好,而在具有大量样本的情况下,神经网络的反演精度有显著提高,而且反演时间比支持向量机少很多.
Support vector machine and neural network theory and internal network training differences of them are studied. Root mean square height and correlation length of Gauss rough surface are inversed by support vector machine and neural network, respectively. Simulation results and inversing errors show that in the case of small numbers of rough surface sample inversion of support vector machine are better than that of neural network, while in the case of sufficient numbers of rough surface samples inversion accuracy of neural network increases and time of inversion by neural network is much less than that of support vector machine.
出处
《计算物理》
CSCD
北大核心
2014年第1期75-84,共10页
Chinese Journal of Computational Physics
基金
国家杰出青年科学基金(61225002)
高等学校博士学科点专项科研基金(20100203110016)资助项目
关键词
支持向量机
粗糙面
神经网络
均方根高度
相关长度
反演
support vector machine
rough surface
neural network
root mean square height
correlation length
inversion
