A STUDY ON AN ACCURATE UNDERWATER LOCATION OF HYBRID UNDERWATER GLIDERS USING MACHINE LEARNING

Abstract

A hybrid underwater glider (HUG) is marine observation equipment that consumes a small amount of energy and offers greater range and navigation times. To achieve reduced energy consumption, however, the HUG uses imprecise navigation sensors, such as mems-type GPS and AHRS, resulting in inaccurate coordination. This study makes a new attempt on the application of machine learning algorithms in a way that complements sensor data errors to improve navigation performance. The proposed algorithm was used to a simulation of the HUG's navigation and control system, after which the updated heading angle was decided by using the previous position data and environmental data, such as ocean current and external forces. The learning algorithm was designed using three layers. Also, the Leaky ReLU activation function was used to solve the problems of gradient vanishing and dying ReLU of machine learning. And to improve the learning efficiency, active functions and the number of layers were changed. The simulation results show the excellent performance of the proposed learning algorithm.