LMBP神经网络改进算法的研究  

On Improved Algorithm of LMBP Neural Networks

作　　者：史步海[1] 朱学峰[1] 

机构地区：[1]华南理工大学自动化科学与工程学院,广东广州510641

出　　处：《控制工程》2008年第2期164-167,共4页Control Engineering of China

基　　金：国家自然科学基金资助项目(62074033);广东省科技攻关基金资助项目(2005B10201005)

摘　　要：论述了BP网络中最优秀的算法之一LMBP算法及其推导过程,分析了标准LMBP算法的特点和不足。为了进一步加快标准LMBP算法的收敛速度,提出了变步长θ的改进LMBP算法。通过采用某水厂混凝沉淀过程真实的实验数据和Matlab仿真程序实验,验证了此改进LMBP算法的可行性和有效性。该改进算法进一步加快了LMBP算法的收敛速度,对于采用LMBP算法神经网络的在线计算具有重要的应用参考价值。LMBP algorithm as one of the excellent algorithms of the BP neural network and its demonstrating process are studied. The characteristics and insufficiency of the standard LMBP algorithm are analysed. To speed up the convergence of standard LMBP algorithm, an improved LMBP algorithm with the changeable step-lengths is proposed. Through adopting the actual data from certain waterworks and Matlab simulation experiment, the feasibility and efficiency of the improved LMBP algorithm are approved, The improved algorithm can quicken the LMBP algorithm convergence speed, so for online computing of the LMBP neural network the proposed algorithm has very important values in applications.

关 键 词：LMBP算法 BP网络 最优化算法 混凝沉淀 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]