Modified minimal resource allocating network for ship motion predictive control

A modified minimal resource allocation network (MRAN) is presented for ship motion predictive control application. To represent the trend of system dynamics change, MRAN is modified by incorporating gradient message in learning process. Resulting variable structure network's structure and connection parameters are adjusted on-line. A predictive controller is proposed with the modified MRAN is employed as a multi-step predictor. The predictive controller with improved MRAN predictor is applied to ship course tracking control, simulation results indicate that the predictive controller has better performance than that of traditional PID controller with higher tracking accuracy and less rudder actions.