样本数据预处理对基于BP神经网络的GPS高程拟合的影响_张昊.pdfVIP

312 Vol. 31No. 2 2 0 1 1 4 JOURNAL OF GEODESY AND GEODYNAM ICS Apr. , 2011 : 1671 5942( 2011) 02 0125 04 BP * GPS 张 昊 王琪洁 朱建军 张晓红 ( , 410083) BP GPS, 3, , , 3 , , , GPS, GPS; ; BP ; ; : P207 : A INFLUENCE OF SAMPLE DATA PREPROCESSING ON BP NEURAL NETWORK-BASED GPS ELEVATION FITTING ZhangH ao, W ang Q ijie, Zhu Jianjun and Zhang X iaohong (D ep artm ent of G eom atics Eng ineering, C entral South Univ ers ity, Chang sha 410083) Abstract The necessity of sample data preprocessing for BP neutral network based GPS elevation fitting is an alyzed, and three preprocessingm ethods are given, which are normalization, centralization and standardization. In order to obtain an optmi almodel, a data preprocessing layer isadded in the network topology and then the three dif ferentmethods are applied in training the neutral networks. Through the comparative analysis of the results derived w ith different preprocessing methods andw ith quadratic surface model, it is found that the influence of three data preprocessing methods on BP neutral newt ork based GPS elevation fitting is different, and the accuracy w ith BP neutral networkmethod is higher than thatw ith quadratic surfacemode.l Key wor s: GPS; elevation fitting; BP neural network; data preprocessing; precision analysis 3: 1 引言 [ 1] GPS GPS , [ 2] , GPS , , , ,