Abstract: In the prediction of slope displacement sequence， there is no specific conclusions on the number of neurons and thresholds in the Elman network hidden layer. The convergence speed is slow， and it is easy to fall into the local solution. Based on this， the improved AFSA-Elman slope displacement prediction network was established by combining the artificial fish swarm algorithm with the Elman network. In order to improve the prediction accuracy and convergence speed of Elman network， the step size of artificial fish swarm algorithm was modified and the initial weights and thresholds of Elman network were optimized by using the powerful optimization ability of the improved fish swarm algorithm. The improved AFSA-Elman network was compared with the traditional Elman network and AFSA-BP network， and the iterative process of the three networks was simulated. The result found that the improved AFSA-Elman prediction network has higher precision and convergence speed than those of the above two prediction network， and it is more suitable for the prediction of slope displacement.

Key words: artificial fish swarm algorithm, Elman network, slope, neural network, displacement