Multi step prediction method of ship pitch based on Conv-Informer model

When ship is sailing at sea, ship motion is irregular and unstable due to the changeable weather and marine environment. If ship pitch is excessive, there will be a large potential safety hazard in the navigation of the ship. The ship motion attitude data has non-linear relationship with the pitch. Single deep learning model cannot simultaneously capture the local spatial characteristics and long-range dependency between multi-variable inputs, so the multi-step prediction accuracy is lower. To improve the accuracy, the convolution network and Informer model are used to build Conv-Informer model for ship pitch prediction. The Conv-Informer model uses convolution structure to extract local spatial information and Informer model to extract long-range dependency of the ship motion data. The simulation results show that the MAPE (Mean Absolute Percentage Error) index of Conv-Informer model decreased by at least 47.59%; the RMSE (Root Mean Square Error) index reduced by at least 30.99%; the MAE (Mean Absolute Error) index decreased by at least 27.86% compared with other models. And the proposed Conv-Informer model can still maintain a lower error in the second and third step of the prediction, which verifies the effectiveness of the Conv-Informer model in multi-step prediction.