The TCN-LSTM Deep Learning Model for Real-Time Prediction of Ship Motions

Real-time and accurate ship motion prediction is of great significance for safe navigation and maritime decision-making. The difficulty lies in the nonlinearity and randomness of ship motion. The long short-term memory model has a strong nonlinear processing capability, which provides a potential way. However, the shortcoming of LSTM makes it difficult to achieve long-term dependencies motion prediction tasks. In present study, a hybrid model by combination of temporal convolutional network and long short-term memory (TCN-LSTM) is proposed, aiming at improving the performance of LSTM model. Based on the ship motion data collected in the Bohai Sea, the performance of the new model is analyzed. It can be found that the TCN-LSTM model can solve longer temporal durations prediction problem than LSTM model. Moreover, the TCN-LSTM model has good applicability under different ship navigation states, and the prediction accuracy is higher with a maximum improvement of 53.43%. Meanwhile, the performance of TCN-LSTM model under different input-output strategies are investigated. Numerical results show that TCN-LSTM has good prediction ability in both strategies, and the prediction accuracy is higher than 99% at the error level of 10%. The TCN-LSTM model has good superiority and applicability for nonlinear and random ship motions prediction.