Computationally efficient MPC for path following of underactuated marine vessels using projection neural network

Abstract

A practical model predictive control (MPC) for path following of underactuated marine vessels, which is a representative marine application, is presented in this paper. Taking advantage of the capability of dealing with multivariable system and input saturation, the MPC method is used to transform the underactuated control problem into the optimization problem with incorporation of input (rudder) constraints. Considering the implementation obstacle of solving optimization problem formulated by the MPC method efficiently, the projection neural network, which is known as parallel computational capability, is employed here to improve the computational efficiency. The full information of ship motion is normally difficult to obtain directly due to the lack of enough measurements; therefore, the state observer is also included. A simple linear model represented the main dynamics of path following of underactuated marine vessels is conceived as predictive (control design) model; meanwhile, in order to demonstrate the effectiveness of proposed control design, all the comparative studies are conducted on a nonlinear high-fidelity simulation model. The simulation results validate that the proposed control design is effective and efficient.