基于BP神经网络的对流层延迟预测研究

STUDY ON PREDICTION OF ZENITH TROPOSPHERIC DELAY BY USE OF BP NEURAL NETWORK

将Levenberg-Marquart算法改进的BP神经网络技术用于GPS对流层延迟的预测,以南加州GPS网数据为例的实验表明:76%的预测值与实际值之间的偏差在3 cm以内;利用BP神经网络进行预测对流层延迟的精度达到厘米级,部分站点达到毫米级。对部分预测效果不佳的站点进行分析,站点空间位置选取不合理、站点海拔与周边训练样本的站点海拔差异较大是造成预测效果不佳的原因。

BP neural network technology improved according to Levenberg-Marquart theory was used for GPS tropospheric delay prediction with the data of Southern California Integrated GPS Network.It is shown that the deviations between the predicted value of 76% GPS Stations and the actual value are less than 3cm.The accuracy of the prediction achieves centimeter level,at some sites amounted to millimeters.Unreasonabl spatial location of the site and site elevation different to the surrounding sites of training samples are the reason which leads to poor prediction for few sites.