BP神经网络转换GPS高程的若干问题探讨  

Discussion on GPS Height Conversion hy A rtificial Neural Network Method

作　　者：吴良才[1] 高宁[1] 

机构地区：[1]东华理工学院地球科学与测绘工程学院,江西抚州344000

出　　处：《工程勘察》2008年第2期46-48,60,共4页Geotechnical Investigation & Surveying

基　　金：地球动力学国家重点实验室开放基金项目(LED0506).

摘　　要：BP神经网络用于GPS高程转换是近年来应用的一种新的方法,该方法有着较高的精度,在某些情况下,该方法优于二次曲面拟合方法,所获得的正常高可满足各种大比例尺测图的精度要求,具有一定的实用价值。但它仍然存在某些不足之处,如网络的隐含层和隐含层单元个数选取,参加学习的样本的质量如何影响仿真精度等,本文结合实例分析了上述问题,得到了一些有益结论。GPS height conversion can get high precise with arithmetic of BP neural network.. Tested by actual cases, the conversion of GPS height by means of neural network is better than the fitting method which simulates the Geoid （or Quasi Geoid）. The normal height obtained can satisfy the required precision of various large scale mapping and has certain practice value. But, there still have some problems, such as how to determinate the number of layers for a network and number of neurons for each layer and so on. This paper discusses the mentioned problems through case studies, and draws some conclusions on the network configuration of BP neural network applied in GPS height conversion.

关 键 词：神经网络 GPS高程 正常高 高程异常 高程转换 

分 类 号：P228.4[天文地球—大地测量学与测量工程]