Abstract: According to the shortcomings of slow convergence of the standard BP （Back-Propagation） algorithm, its main causes are analyzed and the improved algorithm of BP algorithm with the partial counter propagation network is proposed. Through increasing steepness factor of activation functions and increasing coordinator to modify weights when the error is in the back propagation, a full back-propagation type network becomes local-style back-propagation network by means of the analysis of the sensitivity of the network, so as to achieve the purpose of enhancing the learning speed and accuracy. Improved BP algorithm is applied to identification of the stable platform on steerable drilling system. Simulation comparison results show that, in a given accuracy requirement and its convergence speed, this improved BP neural network is superior to traditional BP network, so it has good research and application value.

Key words: Back-Propagation（BP） algorithm, activation function, steepness factor, local-style counter-propagation network, coordinator

摘要： 针对标准BP算法收敛速度慢的缺点，分析了其产生的主要原因，提出了一种改进BP算法。在传统BP算法基础上通过对其激励函数增加陡度因子并在误差反传权值修正时增加协调器，通过对网络灵敏度的分析将全反传式网络变成局部式反传网络，从而达到提高网络学习速率及精度的目的。改进的BP算法应用于导向钻井稳定平台系统的辨识，仿真结果表明该算法收敛速度快，精度高。

关键词: 反向传播（BP）算法, 激励函数, 陡度因子, 局部式反传网络, 协调器