Ship motion model identification based on enhanced Bi-LSTM

Objective Aiming at the low prediction precision and poor adaptability of the ship model based on the data-driven modeling strategy, an enhanced bi-directional long short-term memory （EBLSTM） was proposed for high-precision non-parametric modeling of ships.

Methods Firstly, the feature extraction of bidirectional time dimension is realized by using the feature of Bi-LSTM. On this basis, the spatial dimension features of the convolutional neural network extraction sequence were designed. Finally, multi-head self-attention （MHSA） mechanism is used to deal with the sequence from multiple angles. Using the navigation data of KLVCC2 ships, the prediction effects of EBLSTM model were compared with those of Support Vector Machine （SVM）, Gate Recurrent Unit （GRU） and long short-term memory （LSTM）.

Results The Root Mean Square Error （RMSE） and Mean Absolute Error （MAE） performance indicators of the EBLSTM model in the test set are lower than 0.015 and 0.011, respectively, and the coefficient of determination （R²） is higher than 0.99913. The prediction accuracy is significantly higher than that of SVM, GRU, and LSTM models.

Conclusion The EBLSTM model has excellent generalization performance, excellent prediction stability and precision, and effectively realizes the ship motion model identification.