Self-tuning ship autopilot based on the neural network concept: An empirical approach

Ship autopilots are one of the key factors that guarantee the safety and economical efficiency of marine transportation. In this paper, a self tuning ship autopilot based on neural network concept is introduced. A multilayered feed forward neural network with a fixed part and a tunable part is used for tuning the feedback coefficients of a conventional controller. The connections of tunable part are updated without gradient calculation and iteration. No prior information about controlled object parameters is required. The proposed controller is applied to mathematical models of real ships. The designed autopilot performance is validated for operation cases in calm water and sea wave. The comparison to PID controller and some other control techniques is conducted. The results show that proposed concept may be used for designing ship autopilots and successfully applied for course control of real ships.