Neural network-based depth and horizontal control for autonomous underwater vehicles with prescribed performance

Abstract

This paper presents a study of horizontal and depth controller design for autonomous underwater vehicles (AUVs) regarding model uncertainties, input constraints, and ocean currents. Firstly, the depth-plane and horizontal-plane models of AUV are established comprehensively with ocean currents. Besides, the AUV's characteristics and physical limitations are also described. Secondly, extended perturbation observers (EPOs) are innovatively constructed to estimate ocean currents, which guarantee the accuracy in the case of time-varying ocean currents. To achieve the position errors with prescribed performance guarantees, a finite-time prescribed performance function is proposed such that the convergence time, maximum overshoot, and steady error bounds can be preset directly. Neural network-based adaptive control is adopted to deal with model uncertainties and environmental disturbances. Auxiliary systems with a smooth switching function are introduced to alleviate the effect of input saturation. Furthermore, rigorous theoretical analyses are also carried out to demonstrate the robust stability of the proposed controllers. Finally, extensive numerical simulation studies confirm the strong and conclusive evidence for the proposed method's effectiveness, robustness, anti-jamming ability, and feasibility.