Abstract:
Objectives Polar motion is a critical parameter for the transformation between terrestrial and celestial reference frames. The high-precision polar motion prediction is essential for the application fields such as satellite orbit determination, navigation, and deep space exploration. Traditional polar motion prediction methods, such as least squares (LS) and autoregressive (AR) model, typically assume that annual wobble (AW) and Chandler wobble (CW) have fixed periods. In reality, however, both AW and CW periods exhibit significant time-varying characteristics, which can lead to phase deviations in the medium- to long-term predictions, and ultimately degrade the prediction accuracy.
Methods To enhance the medium- to long-term polar motion predictions, this paper proposes a novel approach by integrating the global optimization capability of differential evolution (DE) algorithm into the classical LS+AR, namely DE+LS+AR. A key advancement is the explicit accommodation of time-varying periods of these two primary geophysical oscillations of AW and CW. First, DE algorithm is employed to adaptively determine the optimal time-varying periods of AW and CW based on a 10-year sliding window, maximizing the merits of LS fitting model. Then, the optimized period parameters are incorporated into the LS model, and combined with AR model for fitting and extrapolation prediction of the polar motion series.
Results The experimental results demonstrate that for 365-day predictions, the accuracy of the proposed DE+LS+AR model is better than 11.60 m(″) and 15.29 m(″) in X and Y components, respectively. Compared with Bulletin A, it represents an improvement of 37.97% and 25.56% in X and Y components, respectively. And compared with the traditional LS+AR model, it represents an improvement of 15.82% and 12.98%, respectively.
Conclusions The DE algorithm effectively extracts and optimizes the non-stationary AW and CW periods of polar motion data. These findings confirm that the proposed novel prediction model can effectively capture the time-varying characteristics of the periodic signals in polar motion data. Thereby it can significantly mitigate the phase drift which occurs in the later stages of traditional LS extrapolation, and markedly improve the accuracy of medium- to long-term prediction.
下载:
