基于LMBP神经网络的涡流传感器曲线拟合研究

Research on eddy current sensor characteristic curve fitting based on neural LMBP network

为了正确反映数字式涡流传感器的实际非线性特性,文中利用Levenberg-Marguardt(L-M)算法优化BP的权值修正量,提出了一种快速收敛的LMBP学习算法,从数学角度详细阐述了LMBP算法的学习过程,并在Matlab环境下设计了具体的网络对数字式涡流传感器特性曲线进行拟合。仿真结果表明,较传统BP网络和传统改进BP网络,LMBP网络有着更快的收敛速度和更强的逼近能力。

In this paper, to accurately reflect the practical characteristics of digital ECS, Levenborg- Marguardt （L-M） algorithm is employed to optimize the correction of BP weight value. A fast convergence LMBP learning algorithm is proposed. The learning process of LMBP algorithm is also illustrated mathematically, and a network is designed via Matlab to fit the ECS characteristic curve. The simulation result shows that LMBP network is superior both in its convergence rate and approximation capability over traditional BP network and improved BP network.