第11卷第1期导航定位学报Vol.11,No.12023年2月JournalofNavigationandPositioningFeb.,2023引文格式:王建敏,毕祥鑫,黄佳鹏.BDS精密钟差短期预报[J].导航定位学报,2023,11(1):30-38.(WANGJianmin,BIXiangxin,HUANGJiapeng.Short-termforecastofprecisionclockdifferenceforBDS[J].JournalofNavigationandPositioning,2023,11(1):30-38.)DOI:10.16547/j.cnki.10-1096.20230105.BDS精密钟差短期预报王建敏,毕祥鑫,黄佳鹏(辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000)摘要:针对传统单一预报模型在钟差预报中误差积累随时间的增加而增大问题,提出一种灰度模型GM(1,1)与长短时记忆神经网络模型(LSTM)的组合模型:采用武汉大学国际全球卫星导航系统服务组织(IGS)数据中心下载的北斗卫星导航系统(BDS)3种轨道不同卫星连续2d的精密钟差数据进行建模,首先用GM(1,1)模型进行预报,然后将GM(1,1)模型的残差利用LSTM神经网络模型进行再次预报;将2种模型的预报结果进行重构,得到最终的预报结果。实验结果表明:GM(1,1)/LSTM组合模型与单一GM(1,1)模型相比,精度提高了60%~89%;GM(1,1)/LSTM组合模型与单一LSTM神经网络相比,精度提升了30%~88%。关键词:钟差预报;灰度模型(GM(1,1));长短时记忆神经网络模型(LSTM);组合模型中图分类号:P228文献标志码:A文章编号:2095-4999(2023)01-0030-09Short-termforecastofprecisionclockdifferenceforBDSWANGJianmin,BIXiangxin,HUANGJiapeng(SchoolofGeomatics,LiaoningTechnicalUniversity,Fuxin,Liaoning123000,China)Abstract:Aimingattheproblemthattheaccumulationoferrorsincreaseswiththeincreaseoftimeforthetraditionalsingleforecastmodelduringtheclockerrorprediction,thepaperproposedacombinationmodelofgraymodelGM(1,1)andlong-shorttermmermorynetwork(LSTM):theprecisionclockdifferencedataoftwoconsecutivedaysfromthreedifferentsatelliteorbitsofBeiDounavigationsatellitesystem(BDS)downloadedbyIGS(InternationalGlobalNavigationSatelliteSystemsService)DataCenteratWuhanUniversityweremodelled,GM(1,1)modelwasusedtoforecastfirstly,andthentheresidualofGM(1,1)modelwasforecastagainbyLSTMneuralnetworkmodel;theforecastresultsbytwomodelswerereconstructedtoobtainthefinalforecastresults.Experimentalresultshowedthatthea...