摘要
针对BP神经网络收敛速度慢和易于陷入局部极小值的问题,采用将遗传算法全局寻优和BP神经网络局部寻优相结合的方法,优化神经网络各层之间的连接权和阈值,提高了BP神经网络的计算精度、收敛速度和泛化能力。本文论述了遗传算法的基本思想、实现过程,并对高程拟合算例进行训练检验,实验结果表明,遗传算法改进的BP神经网络进行GPS高程拟合是可行的,能够有效地提高BP神经网络的拟合精度。
The problems of slow convergence speed and being prone to converge to local minimum are solved by using the method of combining the characteristics of global optimization of GA(Genetic Algorithm) with local optimization of BP neural network.The method optimized the link weight and the threshold of the neural network layers,improved accuracy,convergence speed and generalization ability of the calculation.This paper discusses the basic idea of genetic algorithms,the implementation process,training examples of height fitting.The results of experiments show that BP neural network base on GA is feasible and capable to improve the fitting accuracy effectively in GPS height fitting.
引文
[1]乔仰文等.GPS高程转换的若干问题的研究[J].测绘通报,1999,(11):17~19.
[2]王旭,刘文生.GPS高程拟合方法的研究[J].测绘科学,2010,(S1):28~30.
[3]吴良才,胡振琪.基于神经网络的GPS高程转换方法[J].工程勘察,2004,(2):49~51.
[4]尹爱明,张楚.转换GPS高程的BP神经网络方法研究[J].测绘科学,2008,33(6):78~80.
[5]孔令杰,李捷斌,陈伟.基于遗传算法的BP神经网络在高程拟合中的应用[J].工程勘察,2010,(3):61~64.
[6]丛康林,岳建平,基于SVR的GPS高程拟合模型研究[J].测绘通报,2011,(2):8~11.
[7]Hardy RL.The application of multi quadraic equations and point mass anomaly models to crustal movement studies[J].NOAA Technical Report NOS,1978,76(11).
[8]周明,孙树栋.遗传算法原理及应用[M].北京:国防工业出版社,1999.
[9]王小平,曹立明.遗传算法——理论应用与软件实现[M].西安:西安交通大学出版社,2002.
[10]胡伍生.GPS精密高程测量理论与方法及其应用研究[D].南京:河海大学,2001.