LSTM神经网络模型在GPS变形监测中的应用研究  

Research on Application of LSTM Neural Network Model in GPS Deformation Monitoring

作　　者：李霞 孙茂军 黄永生 Li Xia;Sun Maojun;Huang Yongsheng(1st Institute of Surveying & Mapping Qinghai Province,Xining 810001,China)

机构地区：[1]青海省第一测绘院

出　　处：《甘肃科学学报》2019年第3期24-27,共4页Journal of Gansu Sciences

摘　　要：为提高GPS变形监测在工程应用中的精确度,研究LSTM神经网络在变形监测中的作用。分别利用建立的GM(1,1)模型和LSTM神经网络模型对GPS变形监测工程案例进行应用分析,与GM(1,1)模型相比LSTM神经网络模型预测误差降低幅度可达58%,相对误差降低幅度可达62%,RMSE值降低幅度为66%,结果说明LSTM神经网络模型较GM(1,1)模型有更高的预测精确度,预测结果更接近实际测量结果,深度学习的方法之一LSTM神经网络模型在GPS变形监测中有很高的应用价值。In order to improve the accuracy of GPS deformation monitoring in engineering application,the role of LSTM neural network in deformation monitoring is studied in this paper.The GM(1,1)model and LSTM neural network model are respectively established,analyzing their application in GPS deformation monitoring cases.Compared with GM(1,1)model,the LSTM neural network model is reduced by 58%for forecast error,by 62%for relative error,and by 66%for RMSE value.The results indicate that LSTM neural network model shows a higher accuracy of forecast than GM(1,1)model,and the former's results of forecast are closer to the actual measurement results.LSTM neural network model,one of approaches for deep learning,has a high application value in GPS deformation monitoring.

关 键 词：长短记忆网络 变形监测 循环神经网络 人工智能 

分 类 号：P22[天文地球—大地测量学与测量工程]