Abstract:Precise prediction of GNSS-to-GNSS time offset (GGTO) is critical for multi-GNSS interoperability. To address the limitation that traditional linear prediction methods struggle to accurately characterize the complex nonlinear evolution of the inter-system time offset (GGTO) under the coupling of multiple physical mechanisms, as well as its local micro-noise characteristics, this paper, based on the dynamic evolution mechanism of the time reference, proposes a novel prediction approach that integrates phase space reconstruction (PSR) with support vector regression (SVR). Practical observation data from the national time service center (NTSC) station NT05 are utilized to calculate the inter-system time offsets between GPS, BDS, and Galileo. An SVR model incorporating historical time-series features was constructed and compared with quadratic polynomial (QP) and kalman filter (KF) models. Experimental results demonstrate that, on a one-hour forecasting horizon, the RMSE of the QP and KF models ranges from 1.86 to 2.60 ns and 1.59 to 2.22 ns, respectively, while the proposed SVR model reduces the root mean square errors (RMSE) to 0.65~0.99 ns. Compared to the QP and KF models, the SVR model can improve the prediction accuracy by an average of over 50%. It is concluded that the limitations of traditional linear models in capturing subtle variations of clock offsets are effectively overcome by this research, providing critical technical support for the real-time refinement of integrated navigation and positioning.