Abstract: In order to solve the problem that the clock error prediction caused by falling into the local optimal solution during the training process of BP neural networks,the improved particle swarm optimization BP neural network is used for the clock difference forecast model.Firstly,the method of generating several essential parameters in the particle swarm optimization algorithm is improved.Then the initial weights and thresholds of the BP neural network are used as the positions of the particles.The improved particle swarm optimization algorithm iteratively searches the optimal initial weights and thresholds of the network to improve the clock error prediction of BP neural networks,stability and accuracy.This paper analyzes the improvement principle and uses this model to predict clock error,which proves the effectiveness of the optimization of the algorithm after analyzing the global optimal fitness curve and the experiments of the BP model before and after the particle swarm optimization many times,forecasting the clock difference.Compared with the traditional forecasting models such as the ARMA model and GM(1,1) mode,the accuracy of the clock difference forecast based on the improved particle swarm optimization neural network model is improved by 86.5% and 79%,respectively.

Key words: Clock difference forecast, Particle swarm optimization algorithm, BP neural network