Hybrid Model Identification and Prediction for Docking and Crabbing Motions of a Leisure Boat

Accurate and reliable maneuvering models for leisure boats are crucial for the development of safe navigation and automatic docking solutions. Predicting the docking and crabbing motions of leisure boats presents significant challenges due to the nonlinear characteristics of low-speed operations and the intricate propulsion systems. In this paper, maneuvering models for docking and crabbing operations were examined using various system identification (SI) methodologies. These models were identified with three different approaches; white-box, black-box, and gray-box. The white-box-based SI methodology mathematically identifies the maneuvering model, incorporating the hull force models based on modified MMG models and propeller-rudder force models designed to account for the interaction effects between propellers and rudders. The black-box approach develops a data-driven model to predict the maneuvering dynamics of leisure boats. To determine the optimal black-box model, performances of basic recurrent neural networks (RNNs), long short-term memory (LSTM), gated recurrent unit (GRU), combined with additional layers such as convolutional neural networks (CNNs) and Transformer, time distributed layer, were compared. The gray-box model integrates both the white-box and black-box models to leverage the advantages of both the physical and the data-driven models. The results demonstrated that the gray-box model offers the highest prediction accuracy for the maneuvering model, thereby proving the value of a hybrid approach that combines domain knowledge and artificial intelligence in identifying ship maneuvering models.