Data augmentation methods of dynamic model identification for harbor maneuvers using feedforward neural network

Abstract A dynamic model for an automatic berthing and unberthing controller has to estimate harbor maneuvers, which include berthing, unberthing, approaching maneuvers to berths, and entering and leaving the port. When the dynamic model is estimated by the system identification using feedforward neural networks, a large number of tests or trials are required to measure the various motions of harbor maneuvers. However, the amount of data that can be obtained is limited due to the high costs and time-consuming nature of full-scale ship trials. Therefore, this paper introduces data augmentation to improve the generalization performance of dynamic models identified from a limited dataset. This study used slicing and jittering as data augmentation methods and confirmed their effectiveness by numerical experiments using the free-running model tests. Results of numerical experiments demonstrated that slicing and jittering are effective data augmentation methods but could not improve generalization performance for extrapolation states of the original dataset.