Predicting BDS-3's short-term clock bias using the RIME-WNN model

Aiming at the problems that the clock bias prediction model of the Wavelet Neural Network (WNN) is greatly affected by the selection of network parameters, and the Particle Swarm Optimization Wavelet Neural Network is prone to fall into local optima and has insufficient convergence efficiency in clock bias prediction, a short-term clock bias prediction model for BDS-3 based on the Rime Optimization Algorithm (RIME)-optimized Wavelet Neural Network is proposed. Firstly, the specific steps of the WNN model based on the RIME optimization algorithm in clock bias prediction are elaborated in detail. Then, the stability characteristics and training efficiency of the RIME optimization algorithm during the optimization stage are analyzed to determine the population size that suits the characteristics of clock bias data. Finally, using the BDS-3 clock bias data provided by the Wuhan University Data Center, short-term clock bias prediction experiments with durations of 1 h, 3 h, and 6 h are carried out. The experimental results show that in the 6h prediction, the average prediction accuracy of the RIME-WNN model is better than 0.1 ns, which is 93.92 %, 88.35 %, and 48.11 % higher than that of the Quadratic Polynomial model, the Grey Model (GM(1,1)), and the PSO-WNN model, respectively. In addition, when the RIME-WNN model predicts different types of Beidou satellites, the maximum difference in the Root Mean Square Error (RMSE) is relatively smaller, which fully demonstrates that the model has a wide and good accuracy adaptability when predicting various types of Beidou satellites.