Abstract: In order to solve the problem that accuracy of the existing ultra-fast clock error products is too low to meet the real-time PPP technology, an improved polynomial and periodic clock error prediction model is proposed. The model first uses the polynomial and periodic nonlinear function to make a sliding estimation of the clock error data, and then uses the iterative method to naturally correct the random error of the fitting model to realize the prediction and estimation of the satellite clock error. Compared with the common polynomial model, the gray system model and the polynomial and periodic model, the results show that the improved polynomial and periodic model is more suitable for satellite clock error forecasting, and RMS of the forecast results can reach 0.57 ns and the maximum deviation is 1 ns within a day, which is obviously better than the gray system model and the polynomial and periodic model. With the increase of forecasting time, the forecasting accuracy of the polynomial model, the gray system model and the polynomial and periodic model greatly decreases, while the improved polynomial and periodic model does not change significantly, and the forecast result is stable.