Article(id=1322567084891030195, tenantId=1146029695717560320, journalId=1281212831689347082, issueId=1322567020940481144, articleNumber=null, orderNo=null, doi=10.14075/j.jgg.2025.07.250, pmid=null, cstr=null, oa=null, hot=null, price=null, onlineType=0, articleFormat=0, articleType=null, articleTypeStr=research-article, receivedDate=1752508800000, receivedDateStr=2025-07-15, revisedDate=null, revisedDateStr=null, acceptedDate=null, acceptedDateStr=null, onlineDate=1793254165146, onlineDateStr=2026-10-29, pubDate=1776182400000, pubDateStr=2026-04-15, doiRegisterDate=null, doiRegisterDateStr=null, onlineIssueDate=1793254165146, onlineIssueDateStr=2026-10-29, onlineJustAcceptDate=null, onlineJustAcceptDateStr=null, onlineFirstDate=null, onlineFirstDateStr=null, sourceXml=null, magXml=null, createTime=1793254165146, creator=13701087609, updateTime=1793254165146, updator=13701087609, issue=Issue{id=1322567020940481144, tenantId=1146029695717560320, journalId=1281212831689347082, year='2026', volume='46', issue='4', pageStart='392', pageEnd='521', issueExtLink='null', onlineDate='null', pubDate='1776182400000', pubDateStr='2026-04-15', beforeIssueId=null, nextIssueId=null, price=null, status=1, issueComplete=1, articleOrder=1, issueType=-1, specialIssue=null, createTime=1793254149899, creator='13701087609', updateTime=1793254149899, updator='13701087609', preIssue=null, nextIssue=null, articleTotal=null, ext=null, issueFiles=null, downloadFileDto=null}, startPage=473, endPage=480, ext={EN=ArticleExt(id=1322567085075579572, articleId=1322567084891030195, tenantId=1146029695717560320, journalId=1281212831689347082, language=EN, title=A Regional ZTD Prediction Model Based on GPT3 and Improved Attention LSTM-MHA, columnId=null, journalTitle=Journal of Geodesy and Geodynamics, columnName=null, runingTitle=null, highlight=null, articleAbstract=

Aiming at the problems of low accuracy of empirical models and meteorological parameter model accuracy limited by meteorological parameters in commonly used ZTD prediction models, we propose a regional ZTD prediction model based on improved multi-head attention mechanism combined with long short-term memory network (LSTM-MHA). The model is constructed using data from the third-generation global pressure and temperature model (GPT3), along with station latitude, longitude, elevation, day of year, and hour of day. Using ZTD data from 26 continuously operating reference stations (CORS) in Europe in 2023 as reference values, comparative experiments were conducted with the GPT3 model, LSTM model, and CNN model. The results show that the LSTM-MHA model achieves a mean bias close to 0, with root mean square error and mean absolute error of 13.43 mm and 12.4 mm, respectively. The accuracy of LSTM-MHA model has been improved by 60.8%, 55.2%, and 53.3% compared to the GPT3, CNN, and LSTM models. Evaluation of the LSTM-MHA model performance under different distances and prediction lengths reveals that the proposed method performs stably in both sparse station networks and long-term forecasting scenarios, demonstrating its suitability for regional multi-station prediction tasks.

, authors=Chenglin CAI1, Jie ZHAO1, Wenhui GUAN2, *, Yi QIN3, authorsList=Chenglin CAI, Jie ZHAO, Wenhui GUAN, Yi QIN, authorCompany=null, correspAuthors=Wenhui GUAN, authorNote=null, correspAuthorsNote=null, copyrightStatement=All rights reserved. Unauthorized reproduction is prohibited., copyrightOwner=null, extLink=null, articleAbsUrl=null, sourceXml=null, magXml=null, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=null, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=null, mapNumber=null, fund=null), CN=ArticleExt(id=1322567088502325957, articleId=1322567084891030195, tenantId=1146029695717560320, journalId=1281212831689347082, language=CN, title=基于GPT3和改进注意力LSTM-MHA的区域ZTD预测模型, columnId=1281909276545356011, journalTitle=大地测量与地球动力学, columnName=大地测量学, runingTitle=null, highlight=null, articleAbstract=

针对常用的ZTD预测模型中经验模型精度较低, 气象参数模型精度受限于气象参数的问题, 提出一种结合长短期记忆网络的改进多头注意力机制(LSTM-MHA)的区域ZTD预测模型, 基于第三代全球气压和气温模型(GPT3)数据与站点经纬度、高程、年积日、小时进行建模。选取欧洲地区26个连续运行参考站(CORS)2023年的ZTD数据作为参考值, 与GPT3模型、LSTM模型、CNN模型进行对比实验。结果显示, LSTM-MHA模型结果平均偏差接近于0, 均方根误差和平均绝对误差分别为13.43 mm和12.4 mm, 精度相比于GPT3、CNN、LSTM模型分别提高60.8%、55.2%、53.3%。通过评估LSTM-MHA模型在不同距离和不同预测时长下的性能发现, 所提方法在稀疏站点群和长时预测中均表现稳定, 适合区域多站点预测任务。

, authors=蔡成林1, 赵婕1, 关文绘2, *, 秦懿3, authorsList=蔡成林, 赵婕, 关文绘, 秦懿, authorCompany=null, correspAuthors=关文绘, authorNote=

蔡成林, 博士, 教授, 博士生导师, 主要从事卫星导航与无线通信研究, E-mail:

, correspAuthorsNote=
关文绘, 博士生, 主要从事实时精密单点定位研究, E-mail:
, copyrightStatement=版权所有,未经授权,不得转载。, copyrightOwner=大地测量与地球动力学编辑部, extLink=null, articleAbsUrl=null, sourceXml=TQGzNgGZmhZXidv0tM/xtw==, magXml=q+jwhs/nEZi8jF7i10myAA==, pdfUrl=null, pdf=null, pdfFileSize=null, pdfExtLink=null, richHtmlUrl=null, mobilePdfUrl=null, reviewReport=null, pdfFirstPage=null, abstractGraph=Er08dBIfuDttDU7yPfjHiQ==, abstractGraphContent=null, abstractVideo=null, citation=null, cebUrl=null, magXmlContent=zTWwedHvybA1Po7xocyXAg==, mapNumber=null, fund=null)}, authors=[Author(id=1322567088955310800, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=chengcailin@126.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322567089039196882, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567088955310800, language=EN, stringName=Chenglin CAI, firstName=Chenglin, middleName=null, lastName=CAI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089106305747, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567088955310800, language=CN, stringName=蔡成林, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 湘潭大学自动化与电子信息学院, 湘潭, 411105, bio={"content":"

蔡成林, 博士, 教授, 博士生导师, 主要从事卫星导航与无线通信研究, E-mail:

"}, bioImg=null, bioContent=

蔡成林, 博士, 教授, 博士生导师, 主要从事卫星导航与无线通信研究, E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088695263942, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088703652551, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088712041160, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 湘潭大学自动化与电子信息学院, 湘潭, 411105)])]), Author(id=1322567089169220309, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322567089248912087, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089169220309, language=EN, stringName=Jie ZHAO, firstName=Jie, middleName=null, lastName=ZHAO, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089332798168, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089169220309, language=CN, stringName=赵婕, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 湘潭大学自动化与电子信息学院, 湘潭, 411105, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088695263942, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088703652551, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088712041160, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 湘潭大学自动化与电子信息学院, 湘潭, 411105)])]), Author(id=1322567089412489946, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=gguanwenhui@163.com, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1322567089487987420, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089412489946, language=EN, stringName=Wenhui GUAN, firstName=Wenhui, middleName=null, lastName=GUAN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2 School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089550901981, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089412489946, language=CN, stringName=关文绘, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2 湘潭大学数学与计算科学学院, 湘潭, 411105, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088774955721, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088783344330, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088774955721, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088791732939, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088774955721, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 湘潭大学数学与计算科学学院, 湘潭, 411105)])]), Author(id=1322567089630593759, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322567089697702625, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089630593759, language=EN, stringName=Yi QIN, firstName=Yi, middleName=null, lastName=QIN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3 Chengdu Guoxing Communication Co Ltd, Chengdu 611731, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089764811490, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089630593759, language=CN, stringName=秦懿, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3 成都国星通信有限公司, 成都, 611731, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088879813324, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088888201933, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088879813324, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Chengdu Guoxing Communication Co Ltd, Chengdu 611731, China), AuthorCompanyExt(id=1322567088896590542, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088879813324, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 成都国星通信有限公司, 成都, 611731)])])], keywords=[Keyword(id=1322567089865474787, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, orderNo=1, keyword=tropospheric delay), Keyword(id=1322567089932583652, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, orderNo=2, keyword=multi-head attention mechanism), Keyword(id=1322567090008081125, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, orderNo=3, keyword=long short-term memory network), Keyword(id=1322567090083578598, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, orderNo=4, keyword=regional modeling), Keyword(id=1322567090159076071, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, orderNo=5, keyword=meteorological parameter model), Keyword(id=1322567090230379240, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, orderNo=1, keyword=对流层延迟), Keyword(id=1322567090301682409, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, orderNo=2, keyword=多头注意力机制), Keyword(id=1322567090360402666, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, orderNo=3, keyword=长短期记忆网络), Keyword(id=1322567090448483051, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, orderNo=4, keyword=区域建模), Keyword(id=1322567090515591916, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, orderNo=5, keyword=气象参数模型)], refs=[Reference(id=1322567093900395266, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2025, volume=63, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=1, rfOrder=0, authorNames=Wang D, Wang L K, Kutterer H, journalName=IEEE Transactions on Geoscience and Remote Sensing, refType=null, unstructuredReference= Wang D , Wang L K , Kutterer H . An Advanced Tropospheric Delay Model Based on Gaussian Mixed Long Short-Term Memory Network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63, articleTitle=An Advanced Tropospheric Delay Model Based on Gaussian Mixed Long Short-Term Memory Network, refAbstract=null), Reference(id=1322567093984281347, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2023, volume=97, issue=4, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=2, rfOrder=1, authorNames=Lu C X, Zheng Y X, Wu Z L, journalName=Journal of Geodesy, refType=null, unstructuredReference= Lu C X , Zheng Y X , Wu Z L , et al. TropNet: A Deep Spatiotemporal Neural Network for Tropospheric Delay Modeling and Forecasting[J]. Journal of Geodesy, 2023, 97 (4), articleTitle=TropNet: A Deep Spatiotemporal Neural Network for Tropospheric Delay Modeling and Forecasting, refAbstract=null), Reference(id=1322567094051390212, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2014, volume=57, issue=10, pageStart=3140, pageEnd=3153, url=null, language=null, rfNumber=3, rfOrder=2, authorNames=赵静旸, 宋淑丽, 陈钦明, journalName=地球物理学报, refType=null, unstructuredReference=赵静旸, 宋淑丽, 陈钦明, 等. 基于垂直剖面函数式的全球对流层天顶延迟模型的建立[J]. 地球物理学报, 2014, 57 (10): 3140- 3153, articleTitle=基于垂直剖面函数式的全球对流层天顶延迟模型的建立, refAbstract=null), Reference(id=1322567094114304773, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2014, volume=57, issue=10, pageStart=3140, pageEnd=3153, url=null, language=null, rfNumber=3, rfOrder=3, authorNames=Zhao Jingyang, Song Shuli, Chen Qinming, journalName=null, refType=null, unstructuredReference= Zhao Jingyang , Song Shuli , Chen Qinming , et al. Establishment of a New Global Model for Zenith Tropospheric Delay Based on Functional Expression for Its Vertical Profile[J]. Chinese Journal of Geophysics, 2014, 57 (10): 3140- 3153, articleTitle=null, refAbstract=null), Reference(id=1322567094198190854, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2022, volume=27, issue=1, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=4, rfOrder=4, authorNames=Huang L K, Zhu G, Peng H, journalName=GPS Solutions, refType=null, unstructuredReference= Huang L K , Zhu G , Peng H , et al. An Improved Global Grid Model for Calibrating Zenith Tropospheric Delay for GNSS Applications[J]. GPS Solutions, 2022, 27 (1), articleTitle=An Improved Global Grid Model for Calibrating Zenith Tropospheric Delay for GNSS Applications, refAbstract=null), Reference(id=1322567094273688327, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2013, volume=40, issue=6, pageStart=1069, pageEnd=1073, url=null, language=null, rfNumber=5, rfOrder=5, authorNames=Lagler K, Schindelegger M, Böhm J, journalName=Geophysical Research Letters, refType=null, unstructuredReference= Lagler K , Schindelegger M , Böhm J , et al. GPT2: Empirical Slant Delay Model for Radio Space Geodetic Techniques[J]. Geophysical Research Letters, 2013, 40 (6): 1069- 1073, articleTitle=GPT2: Empirical Slant Delay Model for Radio Space Geodetic Techniques, refAbstract=null), Reference(id=1322567094344991496, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2015, volume=19, issue=3, pageStart=433, pageEnd=441, url=null, language=null, rfNumber=6, rfOrder=6, authorNames=Böhm J, Möller G, Schindelegger M, journalName=GPS Solutions, refType=null, unstructuredReference= Böhm J , Möller G , Schindelegger M , et al. Development of an Improved Empirical Model for Slant Delays in the Troposphere (GPT2w)[J]. GPS Solutions, 2015, 19 (3): 433- 441, articleTitle=Development of an Improved Empirical Model for Slant Delays in the Troposphere (GPT2w), refAbstract=null), Reference(id=1322567094412100361, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2018, volume=92, issue=4, pageStart=349, pageEnd=360, url=null, language=null, rfNumber=7, rfOrder=7, authorNames=Landskron D, Böhm J, journalName=Journal of Geodesy, refType=null, unstructuredReference= Landskron D , Böhm J . VMF3/GPT3: Refined Discrete and Empirical Troposphere Mapping Functions[J]. Journal of Geodesy, 2018, 92 (4): 349- 360, articleTitle=VMF3/GPT3: Refined Discrete and Empirical Troposphere Mapping Functions, refAbstract=null), Reference(id=1322567094491792138, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=8, rfOrder=8, authorNames=null, journalName=null, refType=null, unstructuredReference=Collins J P, Langley R B. The Residual Tropospheric Propagation Delay: How Bad Can It Get [C]. International Technical Meeting of the Satellite Division of the Institute of Navigation, Nashville, 1998, articleTitle=null, refAbstract=null), Reference(id=1322567094550512395, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2001, volume=54, issue=1, pageStart=37, pageEnd=55, url=null, language=null, rfNumber=9, rfOrder=9, authorNames=Penna N T, Dodson A H, Chen W, journalName=Journal of Navigation, refType=null, unstructuredReference= Penna N T , Dodson A H , Chen W . Assessment of EGNOS Tropospheric Correction Model[J]. Journal of Navigation, 2001, 54 (1): 37- 55, articleTitle=Assessment of EGNOS Tropospheric Correction Model, refAbstract=null), Reference(id=1322567094621815564, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2022, volume=13, issue=1, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=10, rfOrder=10, authorNames=Li S, Xu T H, Xu Y, journalName=Atmosphere, refType=null, unstructuredReference= Li S , Xu T H , Xu Y , et al. Forecasting GNSS Zenith Troposphere Delay by Improving GPT3 Model with Machine Learning in Antarctica[J]. Atmosphere, 2022, 13 (1), articleTitle=Forecasting GNSS Zenith Troposphere Delay by Improving GPT3 Model with Machine Learning in Antarctica, refAbstract=null), Reference(id=1322567094718284557, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2024, volume=44, issue=7, pageStart=693, pageEnd=697, url=null, language=null, rfNumber=11, rfOrder=11, authorNames=李金羽, 余学祥, 魏民, journalName=大地测量与地球动力学, refType=null, unstructuredReference=李金羽, 余学祥, 魏民, 等. 基于GPT3和PSO-BPNN的欧洲地区天顶对流层延迟模型[J]. 大地测量与地球动力学, 2024, 44 (7): 693- 697, articleTitle=基于GPT3和PSO-BPNN的欧洲地区天顶对流层延迟模型, refAbstract=null), Reference(id=1322567094802170638, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2024, volume=44, issue=7, pageStart=693, pageEnd=697, url=null, language=null, rfNumber=11, rfOrder=12, authorNames=Li Jinyu, Yu Xuexiang, Wei Min, journalName=null, refType=null, unstructuredReference= Li Jinyu , Yu Xuexiang , Wei Min , et al. Zenith Tropospheric Delay Model in European Region Based on GPT3 and PSO-BPNN[J]. Journal of Geodesy and Geodynamics, 2024, 44 (7): 693- 697, articleTitle=null, refAbstract=null), Reference(id=1322567094886056719, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2022, volume=42, issue=9, pageStart=951, pageEnd=956, url=null, language=null, rfNumber=12, rfOrder=13, authorNames=刘江涛, 刘双童, 叶正真, journalName=大地测量与地球动力学, refType=null, unstructuredReference=刘江涛, 刘双童, 叶正真, 等. 一种区域精密天顶对流层延迟组合预报模型[J]. 大地测量与地球动力学, 2022, 42 (9): 951- 956, articleTitle=一种区域精密天顶对流层延迟组合预报模型, refAbstract=null), Reference(id=1322567094957359888, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2022, volume=42, issue=9, pageStart=951, pageEnd=956, url=null, language=null, rfNumber=12, rfOrder=14, authorNames=Liu Jiangtao, Liu Shuangtong, Ye Zhengzhen, journalName=null, refType=null, unstructuredReference= Liu Jiangtao , Liu Shuangtong , Ye Zhengzhen , et al. A Regional Precise Zenith Tropospheric Delay Combined Prediction Model[J]. Journal of Geodesy and Geodynamics, 2022, 42 (9): 951- 956, articleTitle=null, refAbstract=null), Reference(id=1322567095049634577, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2024, volume=28, issue=4, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=13, rfOrder=15, authorNames=Hu F X, Sha Z M, Wei P Z, journalName=GPS Solutions, refType=null, unstructuredReference= Hu F X , Sha Z M , Wei P Z , et al. Deep Learning for GNSS Zenith Tropospheric Delay Forecasting Based on the Informer Model Using 11-Year ERA5 Reanalysis Data[J]. GPS Solutions, 2024, 28 (4), articleTitle=Deep Learning for GNSS Zenith Tropospheric Delay Forecasting Based on the Informer Model Using 11-Year ERA5 Reanalysis Data, refAbstract=null), Reference(id=1322567095141909266, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2024, volume=12, issue=4, pageStart=34, pageEnd=44, url=null, language=null, rfNumber=14, rfOrder=16, authorNames=白子仪, 徐莹, 冯健, journalName=导航定位学报, refType=null, unstructuredReference=白子仪, 徐莹, 冯健, 等. 基于机器学习的NWP ZTD长短期预测模型[J]. 导航定位学报, 2024, 12 (4): 34- 44, articleTitle=基于机器学习的NWP ZTD长短期预测模型, refAbstract=null), Reference(id=1322567095209018131, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2024, volume=12, issue=4, pageStart=34, pageEnd=44, url=null, language=null, rfNumber=14, rfOrder=17, authorNames=Bai Ziyi, Xu Ying, Feng Jian, journalName=null, refType=null, unstructuredReference= Bai Ziyi , Xu Ying , Feng Jian , et al. NWP ZTD Long-Term and Short-Term Prediction Model Based on Machine Learning[J]. Journal of Navigation and Positioning, 2024, 12 (4): 34- 44, articleTitle=null, refAbstract=null), Reference(id=1322567095276126996, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2018, volume=99, issue=2, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=15, rfOrder=18, authorNames=Blewitt G, Hammond W C, Kreemer C, journalName=EOS, refType=null, unstructuredReference= Blewitt G , Hammond W C , Kreemer C . Harnessing the GPS Data Explosion for Interdisciplinary Science[J]. EOS, 2018, 99 (2), articleTitle=Harnessing the GPS Data Explosion for Interdisciplinary Science, refAbstract=null), Reference(id=1322567095351624469, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2003, volume=3, issue=null, pageStart=115, pageEnd=143, url=null, language=null, rfNumber=16, rfOrder=19, authorNames=Gers F A, Schraudolph N N, Schmidhuber J, journalName=Journal of Machine Learning Research, refType=null, unstructuredReference= Gers F A , Schraudolph N N , Schmidhuber J . Learning Precise Timing with LSTM Recurrent Networks[J]. Journal of Machine Learning Research, 2003, 3, 115- 143, articleTitle=Learning Precise Timing with LSTM Recurrent Networks, refAbstract=null), Reference(id=1322567095418733334, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=2025, volume=17, issue=4, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=17, rfOrder=20, authorNames=Yuan Z D, Lin X, Xu Y S, journalName=Remote Sensing, refType=null, unstructuredReference= Yuan Z D , Lin X , Xu Y S , et al. The VMD-Informer-BiLSTM-EAA Hybrid Model for Predicting Zenith Tropospheric Delay[J]. Remote Sensing, 2025, 17 (4), articleTitle=The VMD-Informer-BiLSTM-EAA Hybrid Model for Predicting Zenith Tropospheric Delay, refAbstract=null), Reference(id=1322567095502619415, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, doi=null, pmid=null, pmcid=null, year=null, volume=null, issue=null, pageStart=null, pageEnd=null, url=null, language=null, rfNumber=18, rfOrder=21, authorNames=null, journalName=null, refType=null, unstructuredReference=Shaker A, Maaz M, Rasheed H, et al. SwiftFormer: Efficient Additive Attention for Transformer-Based Real-Time Mobile Vision Applications[C]. IEEE/CVF International Conference on Computer Vision(ICCV), Paris, 2023, articleTitle=null, refAbstract=null)], funds=[Fund(id=1322567093791343361, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, awardId=2020YFA0713501, language=CN, fundingSource=国家重点研发计划(2020YFA0713501), fundOrder=null, country=null)], companyList=[AuthorCompany(id=1322567088695263942, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088703652551, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088712041160, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 湘潭大学自动化与电子信息学院, 湘潭, 411105)]), AuthorCompany(id=1322567088774955721, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088783344330, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088774955721, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088791732939, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088774955721, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 湘潭大学数学与计算科学学院, 湘潭, 411105)]), AuthorCompany(id=1322567088879813324, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088888201933, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088879813324, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Chengdu Guoxing Communication Co Ltd, Chengdu 611731, China), AuthorCompanyExt(id=1322567088896590542, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088879813324, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 成都国星通信有限公司, 成都, 611731)])], figs=[ArticleFig(id=1322567090628838125, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 1, caption=Structure of the LSTM-MHA algorithm, figureFileSmall=ZK72jqAG9lmj/LcNU15IhQ==, figureFileBig=Er08dBIfuDttDU7yPfjHiQ==, tableContent=null), ArticleFig(id=1322567090695946990, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图1, caption=LSTM-MHA算法结构, figureFileSmall=ZK72jqAG9lmj/LcNU15IhQ==, figureFileBig=Er08dBIfuDttDU7yPfjHiQ==, tableContent=null), ArticleFig(id=1322567090876302063, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 2, caption=Distribution of experimental stations, figureFileSmall=Sx6XM4jIpLqQQ6uxEEjpjQ==, figureFileBig=ji6d/VpYL5AybzyOGMtGnA==, tableContent=null), ArticleFig(id=1322567090939216624, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图2, caption=实验站点分布, figureFileSmall=Sx6XM4jIpLqQQ6uxEEjpjQ==, figureFileBig=ji6d/VpYL5AybzyOGMtGnA==, tableContent=null), ArticleFig(id=1322567091002131185, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 3, caption=Workflow of ZTD prediction, figureFileSmall=lm6gBLuSfJqlo4UUtWgyHA==, figureFileBig=lYMyNJux2IhGuuKabNnnbA==, tableContent=null), ArticleFig(id=1322567091056657138, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图3, caption=ZTD预测流程, figureFileSmall=lm6gBLuSfJqlo4UUtWgyHA==, figureFileBig=lYMyNJux2IhGuuKabNnnbA==, tableContent=null), ArticleFig(id=1322567091106988787, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 4, caption=Comparison of ZTD sequences predicted by different models, figureFileSmall=AcyZNws/XbuLkO/dTOAAjw==, figureFileBig=nCLeWHVlxZ3hXJ1jCVUZWQ==, tableContent=null), ArticleFig(id=1322567091165709044, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图4, caption=不同模型预测的ZTD序列对比, figureFileSmall=AcyZNws/XbuLkO/dTOAAjw==, figureFileBig=nCLeWHVlxZ3hXJ1jCVUZWQ==, tableContent=null), ArticleFig(id=1322567092843430645, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 5, caption=Error distribution of GPT3, LSTM, CNN, and LSTM-MHA models, figureFileSmall=VVeigbQsNRZsCg672Rnn2g==, figureFileBig=4i4teHWePDTOta3ip3MQXw==, tableContent=null), ArticleFig(id=1322567092923122422, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图5, caption=GPT3、LSTM、CNN、LSTM-MHA模型误差分布, figureFileSmall=VVeigbQsNRZsCg672Rnn2g==, figureFileBig=4i4teHWePDTOta3ip3MQXw==, tableContent=null), ArticleFig(id=1322567092998619895, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 6, caption=Distribution of CPAR site and training sites under different distance ranges, figureFileSmall=1rsA/9VU9DdR8cPNnKCgrw==, figureFileBig=seZ7rhQzJhB+jU8roYEc1g==, tableContent=null), ArticleFig(id=1322567093074117368, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图6, caption=CPAR站点与不同距离范围的训练站点分布, figureFileSmall=1rsA/9VU9DdR8cPNnKCgrw==, figureFileBig=seZ7rhQzJhB+jU8roYEc1g==, tableContent=null), ArticleFig(id=1322567093137031929, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 7, caption=Average RMSE for different prediction lengths, figureFileSmall=e8N/dkeO+YkHYZUDRfA0Iw==, figureFileBig=huvC+cLLQqXZylzAt2occg==, tableContent=null), ArticleFig(id=1322567093216723706, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图7, caption=不同预测时长的平均RMSE, figureFileSmall=e8N/dkeO+YkHYZUDRfA0Iw==, figureFileBig=huvC+cLLQqXZylzAt2occg==, tableContent=null), ArticleFig(id=1322567093288026875, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Fig. 8, caption=Pearson correlation analysis and significance test between the precision of LSTM-MHA model and latitude, longitude, elevation, and distance to sea, figureFileSmall=bX0JYjBDV+4veBrEKAeRFQ==, figureFileBig=Tw7OZKfMX06WMZTDcEizFQ==, tableContent=null), ArticleFig(id=1322567093363524348, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=图8, caption=LSTM-MHA模型精度与纬度、经度、高程、距海距离的皮尔逊相关分析和显著性检验, figureFileSmall=bX0JYjBDV+4veBrEKAeRFQ==, figureFileBig=Tw7OZKfMX06WMZTDcEizFQ==, tableContent=null), ArticleFig(id=1322567093455799037, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Tab. 1, caption=

Comparison of prediction errors of different models

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 bias/mm RMSE/mm MAE/mm
平均 最大 最小 平均 最大 最小 平均 最大 最小
GPT3 -2.17 6.60 -12.30 34.29 39.60 28.00 27.44 31.90 22.00
CNN 1.74 13.97 -4.30 29.95 40.39 22.69 27.71 38.12 20.74
LSTM -0.14 6.22 -5.16 28.78 33.65 22.57 26.42 30.98 20.64
LSTM-MHA -0.12 6.41 -5.34 13.43 21.25 8.66 12.40 19.72 7.26
), ArticleFig(id=1322567093535490814, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=表1, caption=

不同模型预测误差对比

, figureFileSmall=null, figureFileBig=null, tableContent=
模型 bias/mm RMSE/mm MAE/mm
平均 最大 最小 平均 最大 最小 平均 最大 最小
GPT3 -2.17 6.60 -12.30 34.29 39.60 28.00 27.44 31.90 22.00
CNN 1.74 13.97 -4.30 29.95 40.39 22.69 27.71 38.12 20.74
LSTM -0.14 6.22 -5.16 28.78 33.65 22.57 26.42 30.98 20.64
LSTM-MHA -0.12 6.41 -5.34 13.43 21.25 8.66 12.40 19.72 7.26
), ArticleFig(id=1322567093619376895, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=EN, label=Tab. 2, caption=

Comparison of prediction errors of CPAR site under different distances

, figureFileSmall=null, figureFileBig=null, tableContent=
距离 bias/mm RMSE/mm MAE/mm
0~200 km 3.48 8.94 7.89
200~300 km 1.99 14.83 12.89
300~400 km 1.11 21.34 18.95
400~500 km 1.31 23.62 21.20
500~600 km 1.01 24.22 21.62
), ArticleFig(id=1322567093690680064, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, language=CN, label=表2, caption=

在不同距离下CPAR站点预测误差对比

, figureFileSmall=null, figureFileBig=null, tableContent=
距离 bias/mm RMSE/mm MAE/mm
0~200 km 3.48 8.94 7.89
200~300 km 1.99 14.83 12.89
300~400 km 1.11 21.34 18.95
400~500 km 1.31 23.62 21.20
500~600 km 1.01 24.22 21.62
)], attaches=null, journal=Journal(id=1281211523964080133, delFlag=0, nameCn=大地测量与地球动力学, nameEn=Journal of Geodesy and Geodynamics, nameHistory1=null, nameHistory2=null, issn=1671-5942, eissn=null, cn=42-1655/P, coden=null, periodic=0, language=CN, oaType=null, ccby=null, superviseOffice=null, ownerOffice=null, pubOffice=null, editorOffice=null, officeType=null, aims=null, clcCode=null, officeProv=null, officeCity=null, officeAddr=null, officeZip=null, officeEmail=null, officePhone=null, editDirector=null, officeDirector=null, officeDirectorPhone=null, officeStaffNum=null, officeEmpNum=null, coverPicUrl=BeWo6OWOseISH6uPOmc2hA==, journalPrice=null, startedYear=null, abbrevIsoEn=Journal of Geodesy and Geodynamics, journalRemark=null, publicationField=null, createdTime=1783394231074, updatedTime=1788951966472, createdBy=18614031015, updatedBy=13041195026, firstLetterCn=D, firstLetterEn=D, subjectCode=Science, subjectName=null, subjectCodeEn=Science, subjectNameEn=null, picCn=BeWo6OWOseISH6uPOmc2hA==, picEn=YHLfVa8gITbOo0FeIjbLDg==, jcr=null, cjcr=null, exts=[JournalExt(id=1304522356412666698, language=CN, name=大地测量与地球动力学, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1788951966622, updatedTime=1788951966622, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.jgg09.com/journalx_ddcl/authorLogOn.action, submissionEditorUrl=http://www.jgg09.com/journalx_ddcl/editorLogOn.action, submissionReviewUrl=http://www.jgg09.com/journalx_ddcl/expertLogOn.action, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""}), JournalExt(id=1304522356576244555, language=EN, name=Journal of Geodesy and Geodynamics, nameHistory1=null, nameHistory2=null, managedBy=, sponsoredBy=, publishedBy=, editorOffice=, officeProv=null, officeCity=null, officeAddr=, officeZip=, editDirector=, officeDirector=null, officePhone=null, coverPicUrl=null, journalRemark=, submitArticleUrl=null, websiteUrl=, createdTime=1788951966661, updatedTime=1788951966661, createdBy=13041195026, updatedBy=13041195026, submissionGuidelinesUrl=, submissionAuthorUrl=http://www.jgg09.com/journalx_ddcl/authorLogOn.action, submissionEditorUrl=http://www.jgg09.com/journalx_ddcl/editorLogOn.action, submissionReviewUrl=http://www.jgg09.com/journalx_ddcl/expertLogOn.action, submissionCeEditorUrl=, submissionAeEditorUrl=, option={"copyright":""})], databaseList=null, tenantJournalId=1281212831689347082, websiteList=[Website(id=1281216327125799907, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1281212831689347082, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/ddcl/CN, language=CN, createTime=1783395376235, createBy=18614031015, updateTime=1783395439358, updateBy=18614031015, name=大地测量与地球动力学-中文, tplId=1146099689490845704, title=大地测量与地球动力学, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1281216715371548758, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=articleTextType, value=kx, createTime=1783395468800, updateTime=1783395468800, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715346382931, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=banner, value=null, createTime=1783395468794, updateTime=1783395468794, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715400908889, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=grayFlag, value=0, createTime=1783395468807, updateTime=1783395468807, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715333800018, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=logo, value=https://castjournals.cast.org.cn/joweb/ddcl/CN/file/pic?fileId=87adUdwolRb6UqGo0QjlJQ==, createTime=1783395468791, updateTime=1783395468791, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715421880411, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=minRunFlag, value=0, createTime=1783395468812, updateTime=1783395468812, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715363160149, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/ddcl/CN/file/pic, createTime=1783395468798, updateTime=1783395468798, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715409297498, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=silenceFlag, value=0, createTime=1783395468809, updateTime=1783395468809, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715358965844, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_cn_619/, createTime=1783395468797, updateTime=1783395468797, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715384131671, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=themeColor, value=null, createTime=1783395468803, updateTime=1783395468803, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216715392520280, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327125799907, code=themeStyle, value=null, createTime=1783395468805, updateTime=1783395468805, creator=18614031015, updator=18614031015)]), Website(id=1281216327713002474, webName=null, webTitle=null, webDomain=null, webCopyrigh=null, webIpcNo=null, seoTitle=null, seoKeywords=null, seoDescription=null, tenantJournalId=null, journalId=1281212831689347082, journalNameCn=null, journalNameEn=null, grayFlag=null, tenantId=1146029695717560320, platformId=null, journalGroupId=null, journalGroupNameCn=null, journalGroupNameEn=null, type=1, domain=https://castjournals.cast.org.cn/joweb/ddcl/EN, language=EN, createTime=1783395376375, createBy=18614031015, updateTime=1783395435863, updateBy=18614031015, name=大地测量与地球动力学-英文, tplId=1146101810881728533, title=Journal of Geodesy and Geodynamics, delFlag=0, indexPage=/home, props=[WebsiteProps(id=1281216680483328066, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=articleTextType, value=kx, createTime=1783395460482, updateTime=1783395460483, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680453967935, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=banner, value=null, createTime=1783395460475, updateTime=1783395460475, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680521076805, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=grayFlag, value=0, createTime=1783395460491, updateTime=1783395460491, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680437190718, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=logo, value=https://castjournals.cast.org.cn/joweb/ddcl/EN/file/pic?fileId=87adUdwolRb6UqGo0QjlJQ==, createTime=1783395460471, updateTime=1783395460471, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680542048327, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=minRunFlag, value=0, createTime=1783395460496, updateTime=1783395460496, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680474939457, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=picServerUrl, value=https://castjournals.cast.org.cn/joweb/ddcl/EN/file/pic, createTime=1783395460480, updateTime=1783395460480, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680533659718, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=silenceFlag, value=0, createTime=1783395460494, updateTime=1783395460494, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680462356544, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=staticResourcePath, value=https://castjournals.cast.org.cn/joweb/cast_kjdb_en_623/, createTime=1783395460477, updateTime=1783395460477, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680495910979, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=themeColor, value=null, createTime=1783395460485, updateTime=1783395460485, creator=18614031015, updator=18614031015), WebsiteProps(id=1281216680508493892, tenantId=1146029695717560320, journalId=null, journalGroupId=null, siteId=1281216327713002474, code=themeStyle, value=null, createTime=1783395460488, updateTime=1783395460488, creator=18614031015, updator=18614031015)])], journalTitle=大地测量与地球动力学, weixinUrl=null, journalUrl=http://www.jgg09.com/CN/home, iacademicId=null, status=1, seqNo=null, journalTitleEn=Journal of Geodesy and Geodynamics, journalPhotoCn=BeWo6OWOseISH6uPOmc2hA==, journalPhotoEn=YHLfVa8gITbOo0FeIjbLDg==, journalFirstLetter=D, journalRecommend=null, journalNew=null, journalCollection=null, jcrJf=null, cjcrJf=null, jcrJfStr=null, cjcrJfStr=null, submissionFirstDecision=null, sciSubjectClassification=null, casSubjectClassification=null, citeScore=null, totalCitationFrequency=null, icpCode=null, psCode=null, advertisingLicenseCode=null, copyrightInformation=null, country=null, option=, provinceCode=null, provinceName=null, collectFlag=false, interPubPlatform=, interPubPlatformUrl=null), detailUrlCn=https://castjournals.cast.org.cn/joweb/ddcl/CN/10.14075/j.jgg.2025.07.250, detailUrlEn=https://castjournals.cast.org.cn/joweb/ddcl/EN/10.14075/j.jgg.2025.07.250, pdfUrlCn=https://castjournals.cast.org.cn/joweb/ddcl/CN/PDF/10.14075/j.jgg.2025.07.250, pdfUrlEn=https://castjournals.cast.org.cn/joweb/ddcl/EN/PDF/10.14075/j.jgg.2025.07.250, aliStartDate=0, aliEndDate=0, collectionFlag=false, citedCount=null, citedUrl=null, previewStatus=0, delFlag=0, hasFullText=1, orderTime=1776182400000, fullTextJson=null, articleText=null, reference=null)
收藏切换
基于GPT3和改进注意力LSTM-MHA的区域ZTD预测模型
收藏切换
蔡成林 1 , 赵婕 1 , 关文绘 2, * , 秦懿 3
大地测量与地球动力学 | 大地测量学 2026,46(4): 473-480
收起
收藏切换
大地测量与地球动力学 |大地测量学 2026 , 46 (4) : 473 -480
基于GPT3和改进注意力LSTM-MHA的区域ZTD预测模型
全屏
[Author(id=1322567088955310800, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=0, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=chengcailin@126.com, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322567089039196882, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567088955310800, language=EN, stringName=Chenglin CAI, firstName=Chenglin, middleName=null, lastName=CAI, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089106305747, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567088955310800, language=CN, stringName=蔡成林, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 湘潭大学自动化与电子信息学院, 湘潭, 411105, bio={"content":"

蔡成林, 博士, 教授, 博士生导师, 主要从事卫星导航与无线通信研究, E-mail:

"}, bioImg=null, bioContent=

蔡成林, 博士, 教授, 博士生导师, 主要从事卫星导航与无线通信研究, E-mail:

, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088695263942, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088703652551, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088712041160, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 湘潭大学自动化与电子信息学院, 湘潭, 411105)])]), Author(id=1322567089169220309, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=1, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322567089248912087, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089169220309, language=EN, stringName=Jie ZHAO, firstName=Jie, middleName=null, lastName=ZHAO, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089332798168, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089169220309, language=CN, stringName=赵婕, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=1, address=1 湘潭大学自动化与电子信息学院, 湘潭, 411105, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088695263942, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088703652551, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088712041160, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088695263942, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=1 湘潭大学自动化与电子信息学院, 湘潭, 411105)])]), Author(id=1322567089412489946, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=2, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=gguanwenhui@163.com, emailSecond=null, emailThird=null, correspondingAuthor=1, authorType=1, ext={EN=AuthorExt(id=1322567089487987420, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089412489946, language=EN, stringName=Wenhui GUAN, firstName=Wenhui, middleName=null, lastName=GUAN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2 School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089550901981, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089412489946, language=CN, stringName=关文绘, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=2, *, address=2 湘潭大学数学与计算科学学院, 湘潭, 411105, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088774955721, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088783344330, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088774955721, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China), AuthorCompanyExt(id=1322567088791732939, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088774955721, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=2 湘潭大学数学与计算科学学院, 湘潭, 411105)])]), Author(id=1322567089630593759, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, orderNo=3, firstName=null, middleName=null, lastName=null, nameCn=null, orcid=null, stid=null, country=null, authorPic=null, dead=0, email=null, emailSecond=null, emailThird=null, correspondingAuthor=0, authorType=1, ext={EN=AuthorExt(id=1322567089697702625, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089630593759, language=EN, stringName=Yi QIN, firstName=Yi, middleName=null, lastName=QIN, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3 Chengdu Guoxing Communication Co Ltd, Chengdu 611731, China, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null), CN=AuthorExt(id=1322567089764811490, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, authorId=1322567089630593759, language=CN, stringName=秦懿, firstName=null, middleName=null, lastName=null, prefix=null, suffix=null, authorComment=null, nameInitials=null, affiliation=null, department=null, xref=3, address=3 成都国星通信有限公司, 成都, 611731, bio=null, bioImg=null, bioContent=null, aboutCorrespAuthor=null)}, companyList=[AuthorCompany(id=1322567088879813324, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, xref=null, ext=[AuthorCompanyExt(id=1322567088888201933, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088879813324, language=EN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 Chengdu Guoxing Communication Co Ltd, Chengdu 611731, China), AuthorCompanyExt(id=1322567088896590542, tenantId=1146029695717560320, journalId=1281212831689347082, articleId=1322567084891030195, companyId=1322567088879813324, language=CN, country=null, province=null, city=null, postcode=null, companyName=null, departmentName=null, remark=3 成都国星通信有限公司, 成都, 611731)])])]
蔡成林1 , 赵婕1, 关文绘2, * , 秦懿3
作者信息
  • 1 湘潭大学自动化与电子信息学院, 湘潭, 411105
  • 2 湘潭大学数学与计算科学学院, 湘潭, 411105
  • 3 成都国星通信有限公司, 成都, 611731
通讯作者:
关文绘, 博士生, 主要从事实时精密单点定位研究, E-mail:
作者简介:

蔡成林, 博士, 教授, 博士生导师, 主要从事卫星导航与无线通信研究, E-mail:

A Regional ZTD Prediction Model Based on GPT3 and Improved Attention LSTM-MHA
Chenglin CAI1 , Jie ZHAO1, Wenhui GUAN2, * , Yi QIN3
Affiliations
  • 1 School of Automation and Electronic Information, Xiangtan University, Xiangtan 411105, China
  • 2 School of Mathematics and Computational Science, Xiangtan University, Xiangtan 411105, China
  • 3 Chengdu Guoxing Communication Co Ltd, Chengdu 611731, China
出版时间: 2026-04-15 doi: 10.14075/j.jgg.2025.07.250
文章导航
收藏切换

针对常用的ZTD预测模型中经验模型精度较低, 气象参数模型精度受限于气象参数的问题, 提出一种结合长短期记忆网络的改进多头注意力机制(LSTM-MHA)的区域ZTD预测模型, 基于第三代全球气压和气温模型(GPT3)数据与站点经纬度、高程、年积日、小时进行建模。选取欧洲地区26个连续运行参考站(CORS)2023年的ZTD数据作为参考值, 与GPT3模型、LSTM模型、CNN模型进行对比实验。结果显示, LSTM-MHA模型结果平均偏差接近于0, 均方根误差和平均绝对误差分别为13.43 mm和12.4 mm, 精度相比于GPT3、CNN、LSTM模型分别提高60.8%、55.2%、53.3%。通过评估LSTM-MHA模型在不同距离和不同预测时长下的性能发现, 所提方法在稀疏站点群和长时预测中均表现稳定, 适合区域多站点预测任务。

对流层延迟  /  多头注意力机制  /  长短期记忆网络  /  区域建模  /  气象参数模型

Aiming at the problems of low accuracy of empirical models and meteorological parameter model accuracy limited by meteorological parameters in commonly used ZTD prediction models, we propose a regional ZTD prediction model based on improved multi-head attention mechanism combined with long short-term memory network (LSTM-MHA). The model is constructed using data from the third-generation global pressure and temperature model (GPT3), along with station latitude, longitude, elevation, day of year, and hour of day. Using ZTD data from 26 continuously operating reference stations (CORS) in Europe in 2023 as reference values, comparative experiments were conducted with the GPT3 model, LSTM model, and CNN model. The results show that the LSTM-MHA model achieves a mean bias close to 0, with root mean square error and mean absolute error of 13.43 mm and 12.4 mm, respectively. The accuracy of LSTM-MHA model has been improved by 60.8%, 55.2%, and 53.3% compared to the GPT3, CNN, and LSTM models. Evaluation of the LSTM-MHA model performance under different distances and prediction lengths reveals that the proposed method performs stably in both sparse station networks and long-term forecasting scenarios, demonstrating its suitability for regional multi-station prediction tasks.

tropospheric delay  /  multi-head attention mechanism  /  long short-term memory network  /  regional modeling  /  meteorological parameter model
蔡成林, 赵婕, 关文绘, 秦懿. 基于GPT3和改进注意力LSTM-MHA的区域ZTD预测模型. 大地测量与地球动力学, 2026 , 46 (4) : 473 -480 . DOI: 10.14075/j.jgg.2025.07.250
Chenglin CAI, Jie ZHAO, Wenhui GUAN, Yi QIN. A Regional ZTD Prediction Model Based on GPT3 and Improved Attention LSTM-MHA[J]. Journal of Geodesy and Geodynamics, 2026 , 46 (4) : 473 -480 . DOI: 10.14075/j.jgg.2025.07.250
天顶对流层延迟(ZTD)在全球卫星导航系统(GNSS)高精度定位和气象学中至关重要[1-3]。在实际应用中可根据是否需要气象参数将ZTD模型分为2类。一类需要实测气象参数,常见的有Black、Saastamoinen和Hopfield模型[4],但很多地区的GNSS监测站点并未配备气象传感器,气象数据获取困难,因此这类模型在实际应用中受到限制。为了解决该问题,许多国内外学者开始研究无需气象参数的模型,即经验模型,可以通过历史经验数据建立各种影响因素与ZTD的映射关系,从而估算ZTD。经过不断研究,一系列经验模型相继被提出,如全球气压和温度(GPT)系列模型[5-7]、新不伦瑞克大学(University of New Brunswick,UNB)系列模型[8]、欧洲地球静止导航重叠服务(European geostationary navigation overlay service,EGNOS)系列模型[9]等。以上经验模型在对流层延迟建模中得到了广泛应用。然而,由于天顶湿延迟(ZWD)模拟较为复杂,难以用物理方程来描述,这类模型在实际应用时仍面临挑战。
机器学习(ML)技术具有从历史数据集中提取信息,并通过回归分析建模非线性特征的强大能力,近年来被广泛应用于对流层延迟建模中。Li等[10]结合长短期记忆(LSTM)和径向基函数(RBF)神经网络,基于GPT3和LSTM模型的ZTD偏差,利用RBF神经网络建模空间位置与ZTD偏差的关系,建立基于GPT3模型的南极洲区域ZTD预报模型。李金羽等[11]将站点经度、纬度、高程、时间与年均dZTD(GNSS真值与GPT3模型值的ZTD残差)作为输入,利用粒子群优化算法(PSO)优化反向传播神经网络(BP),建立欧洲区域ZTD预报模型。刘江涛等[12]为解决模型参数固定,难以表达ZTD剧烈波动的问题,通过支持向量回归(SVR)在空间域建立位置参数与傅里叶级数参数之间的映射关系,输入时间与位置信息即可直接获取ZTD预测值。Hu等[13]利用2011—2021年欧洲中期天气预报中心(ECMWF)第五代再分析数据(ERA5),基于Informer模型,以72 h历史ZTD作为先验信息,实现了24 h ZTD预报。白子仪等[14]利用ERA5数据和国际GNSS服务组织(IGS)的高精度ZTD数据,基于BP神经网络、支持向量机(SVM)和LSTM神经网络,构建ZTD长、短期预测模型。与ERA5产品计算的ZTD相比,所提模型的长期和短期预测精度均有所提高。
当前,结合神经网络预测ZTD成为研究热点,然而一些简单的神经网络架构不具备捕捉ZTD复杂时空特征的能力。此外,GPT系列模型根据时间和位置信息通过外推或内插即可获得目标站点的ZTD,凭借其计算方便的特点得以广泛应用,但GPT3模型根据历史经验数据进行计算,模型精度有限。因此,本研究基于GPT3模型,提出一种结合LSTM和改进多头注意力机制(multi-head attention,MHA)的算法,旨在实现区域高精度ZTD预测。
本研究中使用的实验数据来自内华达大地测量实验室(NGL),NGL产品的ZTD值精度接近国际全球导航卫星系统服务ZTD的官方标称精度(4 mm)[15],在本研究中将其作为ZTD的参考真值,记为ZTDGNSS。GPT3模型可提供全球范围内的经验气压和温度数据,计算简单,输入时间信息和测站坐标参数,就能获得ZTD估计值,记为ZTDGPT3。
GPT3模型计算的ZTD具有明显的季节性变化规律,并且ZTDGPT3与ZTDGNSS之间的残差dZTD表现出高度相关的时空特性。本文基于GPT3模型对dZTD进行建模,以dZTD作为预测模型的输出值。相比直接输出预测ZTD,这种方式除了能描述周期变化趋势外,当温度、湿度和大气压力等气象因素突变时,也不会因这些变化而在模型训练过程中出现过于异常的数值波动。选择纬度(lat)、经度(lon)、高度(height)、年积日(doy)、小时(hod)和ZTDGPT3作为模型构建的输入参数,输出为dZTD。模型输入和输出的关系表达式如下:
$\mathrm{dZTD}=f\left(\text { doy }, \text { hod }, \text { lat }, \text { lon, height }, \mathrm{ZTD}_{\mathrm{GPT} 3}\right)$
式中,f表示模型算法。
对流层延迟具有周期趋势变化、短期波动及高频非线性等特征,简单的神经网络架构难以准确预测。因此本文提出一种结合LSTM与改进多头注意力机制(MHA)的混合预测模型(LSTM-MHA),该模型能更好地分辨时间序列中的全局与局部信息,从而捕捉到ZTD数据中的短期波动和长期趋势。
LSTM-MHA模型由特征嵌入层、LSTM层和改进的多头注意力机制层组成,具体结构见图 1。
特征提取部分由嵌入层、位置编码、时间编码、Dropout层构成,从原始输入数据中提取时间、空间特征和序列信息,并将这些特征融合起来映射到多维空间中。嵌入层采用一维卷积技术,能够提取ZTD序列中的局部特征,并通过整合相邻时间步长的信息,过滤部分高频噪声和突变,使输出序列变得平滑。考虑到ZTD在时间上具有连续性,对ZTD序列进行位置编码和时间编码,使得输出数据的特征表达能力更强。经过卷积层后,数据被送入Dropout层,这一设计能够有效防止过拟合。
LSTM模型最初是为解决循环神经网络(RNN)在处理长序列时存在的梯度消失问题而设计的,通过引入门控机制,LSTM模型能够控制信息是记忆还是遗忘,但早期输入信息可能会被遗忘,导致在长序列中难以捕捉全局和局部之间的复杂关系[16]。注意力机制可以动态分配权重,使模型能够关注序列中的关键时间步;而多头注意力机制通过多个注意力头可同时关注序列中的不同位置,使模型能够更加注重全局和局部之间的关系,增强对重要信息的捕捉能力。其次,多头注意力机制可并行计算,极大提高了模型的计算效率,在长序列和大数据量的预测中也表现出色。在LSTM-MHA模型中,LSTM模块主要捕捉ZTD时序数据中的长期变化趋势,而多头注意力模块则通过多组注意力机制,使模型能够更加注重全局和局部之间的关系,增强了对关键信息的捕捉能力。为克服LSTM模型的局限,本研究将多头注意力机制与LSTM模型结合起来,使模型能更灵活、更全面地捕捉时间序列中的长期趋势和短期波动。
传统的多头注意力机制通过缩放点积操作,对输入序列的每个数据位置,生成查询向量 Q、键向量K和值向量V,从而计算每个位置的注意力权重。本文采用的注意力机制引入了高效加性注意力(efficient additive attention,EAA)[17-18]的思想,将Q和K广播加和后进行非线性变换,再通过前馈网络获得注意力权重。与传统多头注意力机制相比,改进后的注意力机制结合了加性注意力的思想,使得捕捉非线性特征的能力增强,能有效防止因极端值导致的数值不稳定,在ZTD剧烈波动时也能获得较高的预测精度。通过线性元素广播加和替代矩阵乘法,在一定程度上提升了计算效率。
改进后的注意力机制流程如下:如式(2)所示,对输入特征x ∈ RT×dmodel通过多组线性变换分别生成多个头的Q和K。
$\begin{aligned}& \boldsymbol{Q}_i=\boldsymbol{W}_{q_i}(\boldsymbol{x}) \in \boldsymbol{R}^{T \times d_{\text {attn }}} \\& \boldsymbol{K}_i=\boldsymbol{W}_{k_i}(\boldsymbol{x}) \in \boldsymbol{R}^{T \times d_{\text {attn }}}\end{aligned}$
式中,Wqi和Wki为第i个头的可学习参数;dattn= $\frac{d_{\text {model }}}{H}$为子空间维度, H为注意力头的个数, dmodel为输入特征维度;T为序列长度。
在每个注意力头中将Q和K进行逐元素加和并施加tanh激活函数,通过线性变换得到标量注意力分数:
$\operatorname{attn}_i=\operatorname{softmax}\left(\tanh \left(\boldsymbol{W}_{a_i}\left(\boldsymbol{Q}_i+\boldsymbol{K}_i^{\mathrm{T}}\right)\right)\right)$
式中,Qi+KiT表示广播相加后生成的得分矩阵;Wai∈ Rdattn×1为线性变换层。
多头注意力机制通过将输入序列分割成H个子空间,每个子空间都能独立进行注意力计算,每个注意力头根据注意力权重与原始特征序列x执行加权求和,得到各头的特征提取结果;多个注意力头的输出再通过拼接与线性变换进行整合, 表达式为:
$\begin{gathered}\mathbf { head }_i=\text { attn }_i \cdot \boldsymbol{x} \\\text { output }=\boldsymbol{W}_O \cdot \operatorname{concat}\left(\text { head }_1, \text { head }_2, \cdots, \text { head }_H\right)\end{gathered}$
式中,concat为拼接操作;WO为线性输出的权重矩阵。
为保证数据质量,NGL产品只收集数据连续缺失小于36 h且数据完整性高于90%的站点数据。对于偶尔丢失观测数据的情况,通过插值确保每个站点的时间分辨率为1 h,数据覆盖2023全年。研究区域位于37°~65°N、10°W~35°E之间,大部分位于欧洲平原地区。选择26个站点作为测试站点,使用周围8个站点的数据作为训练数据,其中剔除距离测试站点过近的站点,以免影响分析结果。图 2为各测站地理位置分布情况,其中三角形站点为测试站点,灰色圆点为训练站点,提供训练数据。为保证实验的一致性,所有模型输入时间序列长度取相同值,为72个时间步,即72 h。
模型输出项为测试站的残差预测结果dZTD。将模型预测得到的dZTD与ZTDGPT3相加,即为测站的预测ZTD值。为评估本文所提出的模型对ZTD的预测精度,将其与GPT3模型、LSTM模型、CNN模型所得ZTD值进行对比,使用平均偏差(bias)、均方根误差(RMSE)及平均绝对误差(MAE)评估各模型性能,计算公式为:
$\Delta \text{ZTD}=\text{ZTD}_{\text {GNSS }}-\text{ZTD}_{\text {model }}$
$\text { bias }=\frac{\sum\limits_{i=1}^N \Delta \text { ZTD }}{N}$
$\mathrm{RMSE}=\sqrt{\frac{\sum\limits_{i=1}^N \Delta \mathrm{ZTD}^2}{N}}$
$\mathrm{MAE}=\frac{\sum\limits_{i=1}^N|\Delta \mathrm{ZTD}|}{N}$
式中,ZTDGNSS为NGL产品提供的ZTD值;ZTDmodel为不同模型计算的ZTD值;N为样本数。完整的ZTD预测流程如图 3所示。
为了便于观察,选取4个站点(ALAC、GELL、VFCH、MADR)在不同模型(GPT3、LSTM、CNN、LSTM-MHA) 下对预测时间序列进行可视化,结果如图 4所示,其中GNSS为NGL的ZTD参考真值。
从图 4可以看出,由GPT3模型预测的时间序列非常平滑,能够较好地反映出ZTD的近似总体趋势,但对于一些高频非线性变化无法准确捕捉。LSTM和CNN模型的ZTD曲线与GPT3模型具有同样的趋势特征。相比之下,LSTM-MHA模型结果与GNSS真值更为接近,能更为准确地捕捉到ZTD的高频变化,这表明GPT3、LSTM、CNN模型能反映ZTD的近似趋势变化,但对于快速波动的高频变化难以捕捉,而LSTM-MHA模型在ZTD复杂时序预测任务中表现出较高的准确性和稳定性。
为了比较各模型预测精度,将2023年所有测试站点的精度指标进行统计,结果如表 1所示。由表 1可知,GPT3模型计算的ZTD误差在4个模型中最大,LSTM、LSTM-MHA模型的平均bias都接近于0,CNN模型的bias波动较大,平均bias达到1.74 mm。在RMSE方面,LSTM-MHA模型为13.43 mm,相较于GPT3、CNN、LSTM模型分别减少20.86 mm、16.52 mm、15.35 mm,精度分别提升60.8%、55.2%、53.3%。与RMSE类似,LSTM-MHA模型的MAE误差在各个对比模型中最小。由此可见,LSTM-MHA模型在预测性能上显著优于CNN、LSTM模型以及常用的GPT3模型。
为直观展示不同模型预测的ZTD在空间上的误差分布,绘制GPT3、LSTM、CNN、LSTM-MHA模型的bias、RMSE及MAE分布图,具体如图 5所示。
ZTDGPT3是预测模型的重要输入,因此对GPT3模型计算的ZTD误差指标的空间分布特征进行详细分析。不难发现,RMSE值与站点经度无明显相关性,在bias值方面,各站点多为负值,呈现GPT3模型低估ZTD的规律;从纬度来看,除个别低纬度站点因高程较高而表现出低RMSE的情况,GPT3的RMSE基本呈现随纬度升高而降低的规律。
与GPT3模型相比,CNN、LSTM模型预测的ZTD在bias、RMSE和MAE上都有一定程度的降低,而LSTM-MHA具有最高的精度和稳定性。
训练站点与测试站点之间的距离也会影响ZTD预测精度。在模型训练过程中,若训练站点与测试站点之间的距离太远,训练站点的气象条件、地形特征等因素显著异于测试站点,模型训练时难以学习到有用的特征。若距离太近,测试集与训练集数据分布过于相似,无法体现模型对一般站点预测的准确性。
为了评估所提方法的准确性及研究不同密集程度站点群对预测结果的影响,以CPAR站点为基准测试站点,分别取0~200 km、200~300 km、300~400 km、400~500 km、500~600 km不同距离的站点作为训练站点(站点分布如图 6所示), 以此考察在密集、较密集、中等、稀疏站点群中模型的性能。
为了比较LSTM-MHA模型在不同密集程度站点群中的预测性能,统计不同范围的ZTD预测精度,结果如表 2所示。
从表 2可以看出,LSTM-MHA模型的预测精度随着测试站点距训练站点距离的增加而降低,在一定距离内,RMSE随着站点距离的增加呈现明显的线性增长趋势。当距离范围由0~200 km增加到200~300 km时,RMSE上升较快,从8.94 mm上升至14.83 mm。此时模型仍保持较高精度,RMSE在15 mm以下。当距离范围达到300~400 km时,RMSE上升至21.34 mm。此后,范围每增加100 km,RMSE仅增加不到3 mm,说明此时距离对预测精度的影响逐渐减弱。由此可见,当训练站点与测试站点的距离在300 km范围内时,LSTM-MHA模型预测精度较高。当训练站点与测试站点的距离超过300 km时,距离对精度的影响相对不再明显。
由于ZTD时间序列在短时间内变化平稳,多数神经网络模型能较好地捕捉数据中的短期依赖性,而在长时预测中不确定性显著增大,随着误差的累积,简单的神经网络模型难以捕捉数据长期波动的复杂性和依赖性。本文实验中预测时长为1 h,为进一步研究LSTM-MHA模型在不同时长下的预测精度,对比在6 h、12 h、18 h、24 h时长下所有测试站点的平均RMSE(图 7),以评估所提模型在不同预测时长下的性能表现。
由图 7可知,随着预测时长的增加,模型精度降低,其中LSTM、CNN模型误差较大。LSTM-MHA模型在全时段精度最高,在1 h短时预测下,其RMSE为13.43 mm,当预测时长为24 h时,RMSE为19.01 mm,均在20 mm以下。从1 h到24 h的预测时长中,CNN、LSTM、LSTM-MHA模型误差分别增加11.95 mm、9.72 mm、5.58 mm,可见LSTM-MHA模型误差增长最小,较为稳定。在相同预测时长下,LSTM-MHA模型相较其他2个对比模型,精度约高出15 mm。总体而言,LSTM-MHA模型在不同预测时长下均表现出色,优于LSTM和CNN模型。
为研究空间因素对LSTM-MHA模型预测精度的影响,对实验区域26个测试站点夏季与冬季的RMSE与纬度、经度、站点高程、距海距离进行皮尔逊相关分析和显著性检验,实验结果如图 8所示, 其中红线表示完全相关作为参考,蓝线表示实际拟合。
从图 8可以看出,夏季的RMSE明显高于冬季的RMSE。这种趋势可能是因为夏季水汽活动频繁,ZTD预测的不确定性增加。其中高程与RMSE在夏季具有中等强度的相关性,高程升高通常意味着地形复杂性增加、气压和温度变化更剧烈,模型可能难以捕捉高海拔站点的复杂特征。在未来的研究中可通过扩大研究范围、增加站点数量等方法,更好地分析空间特征对ZTD预测精度的影响。
本文提出一种基于LSTM-MHA的算法,实现了区域ZTD高精度预测,主要结论如下:
1) 通过将测站时空特征(经度、纬度、高程、年积日、小时)和GPT3模型计算所得的ZTD值作为输入,经过模型训练后获取高精度的测站ZTD预测值。在LSTM-MHA模型中,LSTM可实现对时序数据的长短期依赖建模,提取序列中的动态变化。引入改进的多头注意力机制,通过多个并行头增强模型对不同子空间非线性特征的捕捉能力,提升了计算效率,适合长序列中动态变化建模。
2) 选取欧洲区域26个测试站点评估所提模型的性能,与GPT3、CNN、LSTM模型相比,LSTM-MHA模型计算ZTD的RMSE分别降低60.8%、55.2%、53.3%。对该模型在不同距离和不同预测时长下进行对比实验, 结果表明,该模型在长距离和长预测时长下仍然表现良好。
综合来看,LSTM-MHA模型的整体性能优于其他对比模型。但由于区域对流层模型精度受多个因素的影响,如测站分布密集程度、测站之间高程差异、训练数据质量等,因此后续研究可以考虑扩大实验区域、增加测试站点,对这些因素作进一步详细探讨,实现在实时PPP中的应用研究。
  • 国家重点研发计划(2020YFA0713501)
参考文献 引证文献
排序方式:
1
Wang D , Wang L K , Kutterer H . An Advanced Tropospheric Delay Model Based on Gaussian Mixed Long Short-Term Memory Network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63
2
Lu C X , Zheng Y X , Wu Z L , et al. TropNet: A Deep Spatiotemporal Neural Network for Tropospheric Delay Modeling and Forecasting[J]. Journal of Geodesy, 2023, 97 (4)
3
赵静旸, 宋淑丽, 陈钦明, 等. 基于垂直剖面函数式的全球对流层天顶延迟模型的建立[J]. 地球物理学报, 2014, 57 (10): 3140- 3153
Zhao Jingyang , Song Shuli , Chen Qinming , et al. Establishment of a New Global Model for Zenith Tropospheric Delay Based on Functional Expression for Its Vertical Profile[J]. Chinese Journal of Geophysics, 2014, 57 (10): 3140- 3153
4
Huang L K , Zhu G , Peng H , et al. An Improved Global Grid Model for Calibrating Zenith Tropospheric Delay for GNSS Applications[J]. GPS Solutions, 2022, 27 (1)
5
Lagler K , Schindelegger M , Böhm J , et al. GPT2: Empirical Slant Delay Model for Radio Space Geodetic Techniques[J]. Geophysical Research Letters, 2013, 40 (6): 1069- 1073
6
Böhm J , Möller G , Schindelegger M , et al. Development of an Improved Empirical Model for Slant Delays in the Troposphere (GPT2w)[J]. GPS Solutions, 2015, 19 (3): 433- 441
7
Landskron D , Böhm J . VMF3/GPT3: Refined Discrete and Empirical Troposphere Mapping Functions[J]. Journal of Geodesy, 2018, 92 (4): 349- 360
8
Collins J P, Langley R B. The Residual Tropospheric Propagation Delay: How Bad Can It Get [C]. International Technical Meeting of the Satellite Division of the Institute of Navigation, Nashville, 1998
9
Penna N T , Dodson A H , Chen W . Assessment of EGNOS Tropospheric Correction Model[J]. Journal of Navigation, 2001, 54 (1): 37- 55
10
Li S , Xu T H , Xu Y , et al. Forecasting GNSS Zenith Troposphere Delay by Improving GPT3 Model with Machine Learning in Antarctica[J]. Atmosphere, 2022, 13 (1)
11
李金羽, 余学祥, 魏民, 等. 基于GPT3和PSO-BPNN的欧洲地区天顶对流层延迟模型[J]. 大地测量与地球动力学, 2024, 44 (7): 693- 697
Li Jinyu , Yu Xuexiang , Wei Min , et al. Zenith Tropospheric Delay Model in European Region Based on GPT3 and PSO-BPNN[J]. Journal of Geodesy and Geodynamics, 2024, 44 (7): 693- 697
12
刘江涛, 刘双童, 叶正真, 等. 一种区域精密天顶对流层延迟组合预报模型[J]. 大地测量与地球动力学, 2022, 42 (9): 951- 956
Liu Jiangtao , Liu Shuangtong , Ye Zhengzhen , et al. A Regional Precise Zenith Tropospheric Delay Combined Prediction Model[J]. Journal of Geodesy and Geodynamics, 2022, 42 (9): 951- 956
13
Hu F X , Sha Z M , Wei P Z , et al. Deep Learning for GNSS Zenith Tropospheric Delay Forecasting Based on the Informer Model Using 11-Year ERA5 Reanalysis Data[J]. GPS Solutions, 2024, 28 (4)
14
白子仪, 徐莹, 冯健, 等. 基于机器学习的NWP ZTD长短期预测模型[J]. 导航定位学报, 2024, 12 (4): 34- 44
Bai Ziyi , Xu Ying , Feng Jian , et al. NWP ZTD Long-Term and Short-Term Prediction Model Based on Machine Learning[J]. Journal of Navigation and Positioning, 2024, 12 (4): 34- 44
15
Blewitt G , Hammond W C , Kreemer C . Harnessing the GPS Data Explosion for Interdisciplinary Science[J]. EOS, 2018, 99 (2)
16
Gers F A , Schraudolph N N , Schmidhuber J . Learning Precise Timing with LSTM Recurrent Networks[J]. Journal of Machine Learning Research, 2003, 3, 115- 143
17
Yuan Z D , Lin X , Xu Y S , et al. The VMD-Informer-BiLSTM-EAA Hybrid Model for Predicting Zenith Tropospheric Delay[J]. Remote Sensing, 2025, 17 (4)
18
Shaker A, Maaz M, Rasheed H, et al. SwiftFormer: Efficient Additive Attention for Transformer-Based Real-Time Mobile Vision Applications[C]. IEEE/CVF International Conference on Computer Vision(ICCV), Paris, 2023
2026年第46卷第4期
20
7
引用本文
BibTeX
文章信息
doi: 10.14075/j.jgg.2025.07.250
  • 接收时间:2025-07-15
  • 首发时间:2026-10-29
  • 出版时间:2026-04-15
补充材料
相关文章
文章信息
作者
出版历史
  • 收稿日期:2025-07-15
基金
国家重点研发计划(2020YFA0713501)
作者信息
    1 湘潭大学自动化与电子信息学院, 湘潭, 411105
    2 湘潭大学数学与计算科学学院, 湘潭, 411105
    3 成都国星通信有限公司, 成都, 611731

通讯作者:

关文绘, 博士生, 主要从事实时精密单点定位研究, E-mail:
参考文献
分享链接
https://castjournals.cast.org.cn/joweb/ddcl/CN/10.14075/j.jgg.2025.07.250
分享至
全文二维码

扫描看全文

引用本文
BibTeX
本文的引用情况
2种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏