Neural networks control of hybrid-driven underwater glider

Abstract

This paper presents a neural network motion control analysis of a hybrid-driven underwater glider. The hybrid-driven underwater glider is a new breed of underwater platform, which combines the features of a conventional glider and autonomous underwater vehicle (AUV). The neural network controller based on multilayer perceptron has been designed as a predictive control. The design objective is to map the control input as well as achieving the target output. A three-layer network, which has six input nodes (control inputs), six hidden layer nodes, and fourteen output nodes is designed as the forward model architecture. Meanwhile, the inverse model of the network is used for the neural network controller. The simulation demonstrates that the control inputs of the glider motion and the target outputs of the reference model are successfully predicted and achieved. The results show that the glider is stable, and the performance of neural network controller is satisfactory, where the value of accuracy is more than 90%.