Earth orientation parameters prediction based on the hybrid SSA + LS + SVM model. (1st December 2022)

• Record Type: Journal Article
• Authors: Yang, Yuguo; Nie, Wenfeng; Xu, Tianhe; Fang, Zhenlong; Xue, Huijie; Sun, Zhangzhen
• Is Part Of: Measurement science & technology. Volume 33:Number 12(2022)
• Publication Date: 2022-12-01
• Subjects: EOP prediction -- polar motion -- length of day -- singular spectrum analysis -- support vector machine; Physical measurements -- Periodicals; Scientific apparatus and instruments -- Periodicals; Equipment and Supplies -- Periodicals; Science -- instrumentation -- Periodicals; Technology -- instrumentation -- Periodicals; Mesures physiques -- Périodiques; Physical measurements; Scientific apparatus and instruments; Periodicals; 502.87
• DOI: 10.1088/1361-6501/ac8ec6
• Languages: English
• ISSNs: 0957-0233

Abstract

Abstract: The high-precision prediction of Earth orientation parameters (EOPs) is essential for astro-geodynamics, high-precision space navigation and positioning, on-board autonomous orbits determination and deep space exploration. However, the prediction accuracy of existing models is much lower than the estimation accuracy of geodetic technical solutions, which affects certain high-precision real-time users. To improve the prediction accuracy of EOP in short- and long-term period, we propose a hybrid model by combining the singular spectrum analysis (SSA), least squares (LSs) and support vector machine (SVM) in the study. Through SSA algorithm, the deterministic time-varying signal of EOP time series can be more precisely and reasonably detected and modeled. Based on the optimization theory, we reconstruct the EOP sequences using SSA and establish the LS extrapolation model based on the reconstructed series. Then, the residuals from SSA reconstruction and those from the LS model, are used for SVM training and prediction. The results of two-year prediction experiments based on the EOP 14 C04 series show that the proposed hybrid model has significant improvements in polar motion (PM) and length of day (LOD) for different prediction intervals (1–360 d) compared with the LS + autoregression (AR) model. The prediction error for x-component of polar motion (PMX) is reduced by 40.2%, 31.0% and 51.4% while that for y-component of polar motion (PMY) is 22.1%, 23.3% and 55.6% for prediction period of 30, 90 and 180 d respectively. For LOD, the maximum prediction improvement can reach to 53.8% during the predicted 360 d. In addition, the proposed method has better accuracy in mid- and long-term PM( x, y ) predictions compared to the Bulletin A, with a 360 d prediction error of 27.273 and 21.741 mas.