Vessel maneuvering model identification using multi-output dynamic fuzzy neural networks

In this paper, a novel vessel maneuvering model (VMM) based on multi-output dynamic fuzzy neural networks is proposed. Data samples are generated from the vessel maneuvering dynamics based on a group of well established nonlinear differential equations. Reasonably, the vessel dynamics identification becomes recurrent prediction since the desired multi-output states are considered to be dependent on not only system inputs, i.e., rudder deflection and propeller revolution, but also previous states. In this case, a novel multi-output dynamic fuzzy neural network (MDFNN) is proposed to identify vessel maneuvering dynamics from data samples. To be specific, the MDFNN starts with zero fuzzy rule and online recruits fuzzy rules via rule node generation criteria and parameter estimation. In addition, insignificant fuzzy rules would be deleted if the rule significance is less than the predefined threshold. As a consequence, the proposed MDFNN based VMM (MDFNN-VMM) reasonably captures essential maneuvering dynamics since the checking process validates the prediction performance with high accuracy. Finally, comprehensive comparisons are carefully presented to demonstrate that the proposed scheme of system identification for vessel motion dynamics is effective. Comprehensive simulation studies are conducted on typical zig-zag maneuvers. And the simulation results demonstrate that the proposed MDFNN-VMM achieves promising performance in terms of approximation and prediction.