Abstract
Aiming at the problem that the traditional algorithm has large prediction error on motion trajectory and short prediction distance, this paper proposes a GA-Elman-Regularization based neural network method. The GA algorithm has the characteristics of parallel search global optimal solution, which makes up for the shortcomings of static property given by neural network model and the tendency of training algorithm to fall into partial optimal solution, and introduces regularization terms to improve the generalization ability of the network, also improves the prediction accuracy of the network. Comparison of experimental results of motion trajectory prediction by Elman neural network, GA-Elman neural network and GA- Elman-Regularization neural network on semi-physical dataset, the predicted average errors are 1.37%, 0.82% and 0.556%. Experiments show that the optimized algorithm improved the generalization ability of the network and the accuracy of prediction.