Model Reference Adaptive Control based on Neural Network for Depth of an AUV

Abstract

This study proposes a Model Reference Adaptive Control (MRAC) approach based on multilayer perceptron (MLP) neural networks to control the depth of a REMUS Autonomous Underwater Vehicle (AUV) during navigation in the presence of range challenges, including hydrodynamic forces and modelling uncertainties. Therefore, Model Reference Adaptive Control (MRAC) is the appropriate controller for this task. The primary objective of this paper was to ensure adaptive control by using the hyperstability concept and applied it to the linear vertical REMUS AUV model. Furthermore, a new approach was introduced: the neural network model reference adaptive control (NNMRAC), which is a combination of the classic MRAC control with a multilayer perceptron neural network (MLPNN), resulting in enhance the  performance and adaptability of the controller . In addition, stability analysis of the new approach is achieved using a Lyapunov candidate function. The effectiveness and feasibility of both adaptive control strategies on vertical AUV motion were evaluated through a comparative analysis conducted using MATLAB/Simulink. This analysis provides valuable information regarding the advantages and limitations of each approach, which can help inform decisions regarding control techniques for regulating the depth of underwater vehicles.