| For more than 20 years,the accumulated coordinate data of IGS reference station provide abundant basic data for Geodesy and geodynamics research,and also provide solid data support for GNSS coordinate time series research.GNSS coordinate time series contains rich physical information and mathematical laws,which can not only reflect the inherited tectonic movement controlled by the same regional tectonic stress field,but also include the nonlinear changes caused by geophysical effects.The accurate analysis of GNSS coordinate time series is of great significance for building deformation monitoring,climate and meteorological prediction,regional crust deformation and so on.Because the outliers in GNSS coordinate time series will have a great impact on the reliability of observation results and the accuracy of data analysis,especially on the application of high-precision GNSS data;on the other hand,after removing outliers,the missing data will lead to the deviation of the station speed obtained by the reference station.Based on this,this paper carries out interpolation work on the basis of outlier detection to reduce the error The estimation bias caused by weak data missing.Finally,this paper establishes a robust prediction model for GNSS coordinate time series,which can provide theoretical and technical reference for disaster prediction and early warning,and further expand the application value of reference station coordinate series.Under this background,this paper takes the nonlinear GNSS coordinate time series as the main research object,summarizes and expands the traditional outlier detection and correction methods(1)Aiming at the difficulty of local outlier detection caused by the nonstationarity and nonlinearity of GNSS elevation coordinate time series,a new outlier detection method based on the combination of Interquartile Range(IQR)and Local Outlier Factor(LOF)is proposed.The experimental results show that LOF algorithm is more accurate than traditional methods in outlier detection,and has better sensitivity to local outliers;while IQR-LOF method can reasonably select the number of outliers,improve the applicability of single LOF algorithm,and get higher precision modeling data.(2)In order to solve the problems of missing data and outlier correction in GNSS coordinate time series,a new interpolation method based on Prophet model is proposed.Through the design of different data random missing proportion and continuous missing length,the comparative experiment is carried out by using prophet model,Lagrange and cubic spline method.The experimental results show that the prophet model has higher interpolation accuracy and better stability than the traditional methods in GNSS coordinate time series,and has obvious advantages for continuous missing data interpolation.(3)The influence of outliers on the prediction accuracy of GNSS coordinate time series is analyzed.Firstly,LOF algorithm is used to detect outliers in coordinate data,and the detected outliers are removed.Then,prophet model is used to interpolate the vacancy data.Finally,Random Forest model(RF)is used to predict.The experimental results show that the prediction accuracy of the data processed by this method is higher,which objectively reflects the necessity of outlier detection.(4)Aiming at the problems of large fluctuation and many outliers of nonlinear GNSS coordinate time series data,based on the in-depth analysis of the characteristics of Prophet model and random forest model,a prophet-RF combined forecasting model is constructed.The combined model solves the defect of the weak prediction ability of the prophet model to the nonlinear part of the time series,and has strong robustness.The experimental results show that,compared with the single prophet model,the combined model of Prophet-RF can better represent the change trend of elevation direction time series,and get higher accuracy prediction data. |
PDF Full Text Request