BP、Hopfield神经网络在位场反演中的应用比较  

THE APPLICATION COMPARISONS OF BPNN AND HNN IN POTENTIAL FIELD INVERSION

作　　者：张新兵[1,2] 王家林[1,2] 陈冰[1,2] 吴健生[1,2] 

机构地区：[1]同济大学海洋地质国家重点实验室 [2]同济大学海洋与地球科学学院,上海200092

出　　处：《物探化探计算技术》2007年第2期161-166,共6页Computing Techniques For Geophysical and Geochemical Exploration

基　　金：中国科学院边缘海地质重点实验室项目(MSGL0610);同济大学理科基金项目(135021906)

摘　　要：比较了BP、Hopfield二种神经网络模型的特性及其运行机制,分别用于位场反演,还比较了各自在位场反演中的应用效果。结果表明:这二种神经网络模型虽然都可用于位场反演,但由于Hopfield网络缺乏学习能力,不能较好地利用已知地质、地球物理信息而受到限制。而BP神经网络具有较强的学习能力,能从已知的信息中得到有利于解决最优化问题的结论,比Hopfield神经网络更加适合于位场的反演问题。The paper compares the features and mechanism of Back-propagation neural networks(BPNN)with that of Hopfield neural networks(HNN),then their application effects in potential field inversion.The results shows that though the both of the neural network models can be used in the potential field inversion,HNN without learning capability can not make use of the prior geological and geophysical data,while on the contrast,BPNN,with their good learning capability and being able to abstract optimum information from prior data is more suitable for potential field inversion than HNN.

关 键 词：BP神经网络 HOPFIELD神经网络 位场反演 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]