In order to solve the problem that BP neural network is easy to fall into the local optimal solution and the convergence speed is slow when dealing with nonlinear and complex environments, an improved artificial bee colony algorithm is proposed to optimize BP neural network and apply it to short-term prediction of clock deviation. Firstly, from the perspective of enhancing the randomness of the step size, improving the search efficiency and maintaining the diversity of the population, the Lévy flight strategy, the teaching and learning optimization algorithm and the fitness-distance balance mechanism were introduced to improve the artificial bee colony algorithm, which effectively improved the global search ability of the algorithm and avoided falling into the local optimal solution. Secondly, the improved artificial bee colony algorithm is combined with BP neural network to be applied to the short-term prediction of satellite clock deviation, and the corresponding calculation steps are given. Then, the high-precision satellite clock products provided by GFZ are selected to compare and analyze the single-day and multi-day forecasts from the aspects of algorithm efficiency, stability and accuracy, so as to verify the applicability of the model. Finally, the prediction results of the optimization model are verified to be applied in PPP. Compared with the QP, BP and ABC-BP models, the average accuracy is increased by 56.55%, 25.11% and 7.07%, respectively, and the improvement effect of MEO-PHM clock is better than that of MEO-Rb clock. The improved artificial bee colony algorithm and the BP combined model have a more concentrated residual distribution, a median closer to zero, and a smaller extreme value, which has high accuracy and stability in the 6 and 12 h prediction. The accuracy of the combined model in the E, N and U directions was improved by using the forecast clock error sequence to improve the accuracy of the results in the E, N and U directions, which were improved by 42.07%, 31.07%, 41.79% and 45.42%, 50.16%, 46.18% compared with the ABC-BP model and the BP model, respectively.
