Ship Course Tracking Control Using Differential of Log-Sum-Exp Neural Network and Model Predictive Control

Abstract

The Differential of Log-Sum-Exp $(DLSE_{T})$ neural network (NN) is combined with model predictive control (MPC) to perform course tracking control based on data. In the past, classical MPC was used to track a given ship reference course, but the ship model should be precisely known, and the cost of MPC online optimization calculation was high. To tackle these problems data driven DLSET NN is used in this paper to approximate the cost functionals based on course data. Off-line neural network training, and DLSET characteristics can reduce the cost of online optimization, and MPC can ensure that the rudder angle constraint is satisfied. According to the simulation results, the DLSET-based MPC is feasible in ship course tracking control.