Abstract: Aiming at the short-term prediction of the BeiDou-3 Navigation Satellite System (BDS-3) satellite clock, based on the analysis of the frequency stability of the satellite atomic clock, and autoregressive integrated moving average, grey model, linear polynomial and quadratic polynomial four kinds of clock bias prediction models are used to fit and predict based on 30 days of data. The experimental results show that: 1) Compared with BeiDou-2 Navigation Satellite System (BDS-2), the BDS-3 atomic clock has higher stability, among which the thousand-second stability, ten-thousand-second stability and daily stability of the BDS-3 hydrogen clock have reached 4.2×10⁻¹⁴, 1.89×10⁻¹⁴ and 4.14×10⁻¹⁵ respectively; 2) The prediction stability of the BDS-3 hydrogen clock and the BDS-3 new rubidium clock is significantly improved compared to the BDS-2 rubidium clock, and the prediction accuracy of the BDS-3 hydrogen clock is the best, reaching 0.12 ns, 0.18 ns, 0.30 ns at 3 h, 6 h and 12 h respectively; 3) Among the four models, the time series model has the highest forecast accuracy, with accuracy of 0.26 ns, 0.47 ns and 0.96 ns at 3 h, 6 h and 12 h, respectively.