Short-term motion prediction of a semi-submersible platform based on a combined prediction model

The reliable short-term prediction of offshore structure motions holds significant importance for operational safety and decision-making. In this paper, a novel combined prediction model (CPM) designed for short-term motion forecasting of a semi-submersible platform was proposed. The dataset employed in the study was derived from a model test conducted in a wave basin. In the process of sub-model selection, long short-term memory (LSTM), gated recurrent unit (GRU), and temporal convolutional network (TCN) were chosen based on a comprehensive analysis of sub-model evaluation index. Through the integration of a multi-layer feedforward neural network to optimize the weight allocation among sub-model predictions, a secondary prediction was attained, resulting in notably improved prediction accuracy. The results demonstrate that the proposed model exhibits excellent transferability across different wave conditions and outperforms individual models in prediction performance. Combined with the analysis of different decomposition algorithms, it is evident that the variational mode decomposition (VMD) algorithm could conserve computational resources and maximize prediction accuracy to a significant extent. Extending the prediction to encompass the 6 degrees of freedom (DOFs) motion time series of the platform, the proposed model can obtain highly accurate prediction results on all 6-DOF motions.