基于遗传算法的BP神经网络在高程拟合中的应用  

Application of BP neural network based on GA in height fitting

作　　者：孔令杰[1,2] 李捷斌[1,2] 陈伟[3] 

机构地区：[1]长安大学地质工程与测绘学院,西安710054 [2]西部矿产资源与地质工程教育部重点实验室,西安710054 [3]南昌工程学院水利工程系,南昌330099

出　　处：《工程勘察》2010年第3期61-64,共4页Geotechnical Investigation & Surveying

基　　金：国家自然科学基金项目(40672173)

摘　　要：针对传统BP神经网络难以选取具有全局性的初始点的缺点,使用改进的遗传算法全局搜索优化神经网络各层之间的连接权和阈值,提高了BP神经网络的收敛速度和泛化能力。结合高程拟合算例进行训练检验,证明该方法是一种改进BP神经网络的有效方法。Because the weakness of traditional BP neural network is difficuhy to select the initial point, the improved genetic algorithm with global searching is used for optimization of the link weight and the threshold of the neural network layers for improving the capability of traditional BP neural network. By comparison, the convergence rate and generalization ability of BP based on genetic algorithm are higher than that of the traditional BP neural network. In the example of height fitting, it is proved that the improved algorithm is efficient.

关 键 词：BP神经网络 遗传算法 收敛速度 泛化能力 高程拟合 

分 类 号：TU457[建筑科学—岩土工程]