This paper investigates the possibility of the machine learning technique being applicable in the real-time prediction of wave-induced ship motions. An integrated machine learning model is proposed by considering the two different physical attributes in the equation of motion: memory effects of past motion history and excitation forces induced by incident waves. A long short-term memory layer and a single fully connected layer were combined to establish this machine learning model. A database was constructed through the impulse response function-based numerical simulations for various ocean environments. After training, short-term deterministic predictions were conducted for new environments, and the effects of ship motion records were investigated. The response amplitude operators were evaluated based on regular wave simulations. The machine learning model was observed to have successfully learned the seakeeping characteristics of a ship.
Autonomous ships have emerged as an important issue in the transportation market, including the shipbuilding industry. The global market size of the autonomous ship is projected to approximately triple by 2030 (MarketsandMarkets, 2021). The development of next-generation vessels, increasing investments in autonomous ships, and the surge of demand for safety have been the primary factors presenting many new technical issues in marine automation systems (Geertsma et al., 2017). Among these issues, a real-time prediction for ship operation performance is strongly required for the navigation process to improve safety and efficiency (Lee et al., 2022). Accurate and efficient predictions of seakeeping and maneuvering, however, still remain challenging, owing to the expensive computation cost for numerical simulations (Weymouth and Yue, 2013). Existing or conventional methods have difficulty immediately responding to the provided information during a voyage.
