A novel precise SLS+Autoformer prediction model of universal time/length of day with effective angular momentum information

High-precision predictions of Universal Time and Coordinated Universal Time (UT1-UTC) and Length of Day (LOD) play crucial roles in fields such as satellite navigation, astrometry, and deep space exploration. Traditional linear methods suffer from rapid accuracy degradation and difficulty in capturing long-period variations in medium-to-long-term predictions. To improve the accuracy of UT1-UTC and LOD predictions, we propose a novel precise prediction model, namely SLS + Autoformer, that integrates the Segmented Least-squares (SLS) and the deep learning model Autoformer with the Effective Angular Momentum (EAM) and Geodetic Angular Momentum (GAM) data. Experimental results demonstrate that, for UT1-UTC prediction of 30, 60, and 90 days, the mean absolute error (MAE) values reach 2.1917 ms, 5.1096 ms, and 8.3726 ms, respectively. Compared to the IERS Bulletin A product, the accuracy improvements attained were around 40%, 38%, and 32%, respectively. Compared to SLS + AR, the improvements are around 15%, 12%, and 14%, respectively. For LOD prediction of 30, 60, and 90 days, the MAE values are 0.1632 ms, 0.1762 ms, and 0.1869 ms, respectively. Compared to the best results with the traditional method using EAM data in the Second Earth Orientation Parameters Prediction Comparison Campaign (2nd EOP PCC), the accuracy improvements achieved were around 34%, 35%, and 38%, respectively. Compared to SLS + AR, the improvements are around 14%, 11%, and 7%, respectively. The results confirm that the proposed novel precise SLS + Autoformer model can significantly enhance the medium-term prediction accuracy of UT1-UTC and LOD.