An improved particle swarm optimization algorithm applied to long short-term memory neural network for ship motion attitude prediction

Abstract

This paper proposes a prediction method of ship motion attitude with high accuracy based on the long short-term memory neural network. The model parameters should be initialized randomly, resulting in critical decreases of the nonlinear learning ability of current parameter optimization methods. Therefore, a multilayer heterogeneous particle swarm optimization is proposed to optimize the parameters of long short-term memory neural network and applied to the prediction of ship motion. In multilayer heterogeneous particle swarm optimization, this paper proposes the concept of attractors, transforms the speed update equation, enhances the information interaction ability between particles, improves the optimization performance of the particle swarm optimization algorithm, and improves its optimization effect on the parameters of the long short-term memory networks. In the simulations, the measured data were used as input to predict the results of the ship motion. The results showed that the proposed method offers higher learning accuracy, faster convergence speed, and better prediction performance for accurate estimation of ship motion attitude than existing methods.