Machine learning for ship heave motion prediction: Online adaptive cycle reservoir with regular jumps

Ship heave motion prediction is an important part of wave compensation. Machine learning algorithms, such as Long Short-Term Memory (LSTM) networks, have demonstrated commendable predictive capabilities in this area. However, traditional methods rely on offline training, which limits their ability to obtain effective information from new available data, and in practical scenarios, the predictive effectiveness of traditional models deteriorates when ship motion data changes unforeseen. To address this issue, this paper introduces an Online Adaptive Cycle Reservoir with Regular Jumps (CRJ) for predicting ship heave motion. The method activates a Bayesian optimizer when the model prediction deteriorates and adjusts the model parameters online to adapt the current ship motion data. The proposed method has been successfully applied in ship heave motion prediction experiments and compared with three models (CRJ, Deep Delay CRJ, LSTM) the results show that the proposed model exhibits superior stability and robustness. In addition, the online training time of the model is sufficiently short enough to satisfy the demand for online prediction. •An online learning model for ship heave motion prediction updates with evolving data patterns over time.•Bayesian tuning dynamically adjusts model parameters online, improving the model's accuracy and robustness.•Hardware experiments validated the online learning model's effectiveness and robustness.