Barrier Lyapunov function-based adaptive optimized control for full-state and input-constrained dynamic positioning of marine vessels with simulation and model-scale tests

Abstract

For practical marine engineering, ensuring system performance and avoiding violations of safety-related system constraints in various scenarios remains a challenging problem. To tackle this problem, this paper addresses adaptive neural network (NN) optimized control for dynamic positioning (DP) of marine vessels suffering from unknown model uncertainties and control gains, ocean disturbances, and full-state/input constraints. To realize model-free optimized control, the reinforcement learning (RL) with a critic–actor–identifier architecture is constructed on the basis of NN approximation, in which the actor, critic, and identifier NNs are designed to perform control actions, evaluate control performance, and estimate unknown dynamics, respectively. To effectively guarantee the system safety, the critics and actors in the corresponding subsystems are decomposed into a guaranteed safety-related constraint term with respect to the barrier Lyapunov function (BLF) and an unknown dynamic term to be learned. Such a control strategy can ensure that all the state variables are restricted to a predefined safety region, even in the learning control process. From the above logical ideas, the adaptive optimized kinematic and kinetic control laws are presented to guarantee that the whole vessel system is optimized. Moreover, the motion control responses of vessels in various scenarios are realistically simulated through convincing simulations and model-scale tests. The validation results prove the applicability and feasibility of the proposed DP control strategy for practical offshore engineering.